Release v2.3.26

Workflow.yaml now supports all agents.yaml features:

Agent-level features:
- framework (praisonai, crewai, autogen)
- process (sequential, hierarchical, workflow)
- manager_llm for hierarchical process
- function_calling_llm
- max_rpm (rate limiting)
- max_execution_time (timeout)
- reflect_llm, min_reflect, max_reflect
- system_template, prompt_template, response_template
- backstory (alias for instructions)

Step-level features:
- expected_output (like agents.yaml tasks)
- context (task dependencies)
- output_json, output_pydantic (structured output)
- create_directory
- callback
- async_execution

Added 14 new TDD tests for feature parity verification.
All 41 YAML parser tests pass.

.github/actions/claude-code-action/action.yml

@@ -86,13 +86,29 @@ runs:
     - name: Extract GitHub Context and Create Prompt
       shell: bash
       id: prepare_context
+      env:
+        # Pass user-controlled inputs via environment variables to prevent script injection (GHSL-2025-093)
+        EVENT_COMMENT_BODY: ${{ github.event.comment.body }}
+        EVENT_ISSUE_TITLE: ${{ github.event.issue.title }}
+        EVENT_ISSUE_BODY: ${{ github.event.issue.body }}
+        EVENT_ISSUE_NUMBER: ${{ github.event.issue.number }}
+        EVENT_PR_TITLE: ${{ github.event.pull_request.title }}
+        EVENT_PR_NUMBER: ${{ github.event.pull_request.number }}
+        EVENT_REVIEW_BODY: ${{ github.event.review.body }}
+        EVENT_ASSIGNEE_LOGIN: ${{ github.event.assignee.login }}
+        INPUT_TRIGGER_PHRASE: ${{ inputs.trigger_phrase }}
+        INPUT_ASSIGNEE_TRIGGER: ${{ inputs.assignee_trigger }}
+        INPUT_DIRECT_PROMPT: ${{ inputs.direct_prompt }}
+        EVENT_ACTION: ${{ github.event.action }}
+        EVENT_NAME: ${{ github.event_name }}
+        REPO_NAME: ${{ github.repository }}
       run: |
-        echo "🔍 Extracting GitHub context from event: ${{ github.event_name }}"
+        echo "🔍 Extracting GitHub context from event: $EVENT_NAME"
         
         # Function to check for trigger phrase
         check_trigger() {
           local text="$1"
-          local trigger="${{ inputs.trigger_phrase }}"
+          local trigger="$INPUT_TRIGGER_PHRASE"
           if [[ "$text" == *"$trigger"* ]]; then
             return 0
           fi
@@ -104,11 +120,11 @@ runs:
         USER_REQUEST=""
         CONTEXT_INFO=""
         
-        case "${{ github.event_name }}" in
+        case "$EVENT_NAME" in
           "issue_comment")
-            COMMENT_BODY="${{ github.event.comment.body }}"
-            ISSUE_TITLE="${{ github.event.issue.title }}"
-            ISSUE_NUMBER="${{ github.event.issue.number }}"
+            COMMENT_BODY="$EVENT_COMMENT_BODY"
+            ISSUE_TITLE="$EVENT_ISSUE_TITLE"
+            ISSUE_NUMBER="$EVENT_ISSUE_NUMBER"
             
             if check_trigger "$COMMENT_BODY"; then
               TRIGGER_FOUND="true"
@@ -118,9 +134,9 @@ runs:
             ;;
             
           "pull_request_review_comment")
-            COMMENT_BODY="${{ github.event.comment.body }}"
-            PR_TITLE="${{ github.event.pull_request.title }}"
-            PR_NUMBER="${{ github.event.pull_request.number }}"
+            COMMENT_BODY="$EVENT_COMMENT_BODY"
+            PR_TITLE="$EVENT_PR_TITLE"
+            PR_NUMBER="$EVENT_PR_NUMBER"
             
             if check_trigger "$COMMENT_BODY"; then
               TRIGGER_FOUND="true"
@@ -130,9 +146,9 @@ runs:
             ;;
             
           "pull_request_review")
-            REVIEW_BODY="${{ github.event.review.body }}"
-            PR_TITLE="${{ github.event.pull_request.title }}"
-            PR_NUMBER="${{ github.event.pull_request.number }}"
+            REVIEW_BODY="$EVENT_REVIEW_BODY"
+            PR_TITLE="$EVENT_PR_TITLE"
+            PR_NUMBER="$EVENT_PR_NUMBER"
             
             if check_trigger "$REVIEW_BODY"; then
               TRIGGER_FOUND="true"
@@ -142,17 +158,17 @@ runs:
             ;;
             
           "issues")
-            ISSUE_BODY="${{ github.event.issue.body }}"
-            ISSUE_TITLE="${{ github.event.issue.title }}"
-            ISSUE_NUMBER="${{ github.event.issue.number }}"
+            ISSUE_BODY="$EVENT_ISSUE_BODY"
+            ISSUE_TITLE="$EVENT_ISSUE_TITLE"
+            ISSUE_NUMBER="$EVENT_ISSUE_NUMBER"
             
             if check_trigger "$ISSUE_TITLE" || check_trigger "$ISSUE_BODY"; then
               TRIGGER_FOUND="true"
               USER_REQUEST="$ISSUE_BODY"
               CONTEXT_INFO="Issue #$ISSUE_NUMBER: $ISSUE_TITLE"
-            elif [[ "${{ github.event.action }}" == "assigned" && -n "${{ inputs.assignee_trigger }}" ]]; then
-              ASSIGNEE="${{ github.event.assignee.login }}"
-              if [[ "$ASSIGNEE" == "${{ inputs.assignee_trigger }}" ]]; then
+            elif [[ "$EVENT_ACTION" == "assigned" && -n "$INPUT_ASSIGNEE_TRIGGER" ]]; then
+              ASSIGNEE="$EVENT_ASSIGNEE_LOGIN"
+              if [[ "$ASSIGNEE" == "$INPUT_ASSIGNEE_TRIGGER" ]]; then
                 TRIGGER_FOUND="true"
                 USER_REQUEST="$ISSUE_BODY"
                 CONTEXT_INFO="Issue #$ISSUE_NUMBER assigned to $ASSIGNEE: $ISSUE_TITLE"
@@ -162,9 +178,9 @@ runs:
         esac
         
         # Check for direct prompt override
-        if [[ -n "${{ inputs.direct_prompt }}" ]]; then
+        if [[ -n "$INPUT_DIRECT_PROMPT" ]]; then
           TRIGGER_FOUND="true"
-          USER_REQUEST="${{ inputs.direct_prompt }}"
+          USER_REQUEST="$INPUT_DIRECT_PROMPT"
           CONTEXT_INFO="Automated GitHub workflow"
         fi
         
@@ -181,9 +197,9 @@ runs:
         cat > /tmp/claude-action/github-context-prompt.txt << EOF
         You are Claude Code, an AI assistant helping with GitHub workflows and code.
         
-        Repository: ${{ github.repository }}
+        Repository: $REPO_NAME
         Context: $CONTEXT_INFO
-        Event: ${{ github.event_name }}
+        Event: $EVENT_NAME
         
         User Request:
         $USER_REQUEST

.github/workflows/gemini-issue-review.yml

@@ -46,12 +46,16 @@ jobs:
         id: issue_data
         env:
           GH_TOKEN: ${{ steps.generate_token.outputs.token }}
+          # Pass user-controlled inputs via environment variables to prevent script injection (GHSL-2025-093)
+          EVENT_NAME: ${{ github.event_name }}
+          EVENT_ISSUE_NUMBER: ${{ github.event.issue.number }}
+          EVENT_COMMENT_BODY: ${{ github.event.comment.body }}
         run: |
           # Get issue number from event or context
-          if [ "${{ github.event_name }}" = "issue_comment" ]; then
-            ISSUE_NUMBER="${{ github.event.issue.number }}"
-          elif [ "${{ github.event_name }}" = "issues" ]; then
-            ISSUE_NUMBER="${{ github.event.issue.number }}"
+          if [ "$EVENT_NAME" = "issue_comment" ]; then
+            ISSUE_NUMBER="$EVENT_ISSUE_NUMBER"
+          elif [ "$EVENT_NAME" = "issues" ]; then
+            ISSUE_NUMBER="$EVENT_ISSUE_NUMBER"
           else
             echo "Unable to determine issue number"
             exit 1
@@ -62,8 +66,8 @@ jobs:
           
           # Extract additional instructions from comment if triggered by comment
           ADDITIONAL_INSTRUCTIONS=""
-          if [ "${{ github.event_name }}" = "issue_comment" ]; then
-            COMMENT_BODY="${{ github.event.comment.body }}"
+          if [ "$EVENT_NAME" = "issue_comment" ]; then
+            COMMENT_BODY="$EVENT_COMMENT_BODY"
             ADDITIONAL_INSTRUCTIONS=$(echo "$COMMENT_BODY" | sed -n 's/.*@gemini[[:space:]]*\(.*\)/\1/p' | head -1)
           fi
           

.github/workflows/gemini-pr-review.yml

@@ -55,21 +55,27 @@ jobs:
         id: get_pr
         env:
           GITHUB_TOKEN: ${{ steps.generate_token.outputs.token }}
+          # Pass user-controlled inputs via environment variables to prevent script injection (GHSL-2025-093)
+          EVENT_NAME: ${{ github.event_name }}
+          EVENT_INPUTS_PR_NUMBER: ${{ github.event.inputs.pr_number }}
+          EVENT_ISSUE_NUMBER: ${{ github.event.issue.number }}
+          EVENT_PR_NUMBER: ${{ github.event.pull_request.number }}
+          EVENT_COMMENT_BODY: ${{ github.event.comment.body }}
         run: |
-          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
-            PR_NUMBER=${{ github.event.inputs.pr_number }}
-          elif [ "${{ github.event_name }}" = "issue_comment" ]; then
-            PR_NUMBER=${{ github.event.issue.number }}
+          if [ "$EVENT_NAME" = "workflow_dispatch" ]; then
+            PR_NUMBER=$EVENT_INPUTS_PR_NUMBER
+          elif [ "$EVENT_NAME" = "issue_comment" ]; then
+            PR_NUMBER=$EVENT_ISSUE_NUMBER
           else
-            PR_NUMBER=${{ github.event.pull_request.number }}
+            PR_NUMBER=$EVENT_PR_NUMBER
           fi
           
           echo "pr_number=$PR_NUMBER" >> "$GITHUB_OUTPUT"
           
           # Extract additional instructions from comment (if triggered by comment)
           ADDITIONAL_INSTRUCTIONS=""
-          if [ "${{ github.event_name }}" = "issue_comment" ]; then
-            COMMENT_BODY="${{ github.event.comment.body }}"
+          if [ "$EVENT_NAME" = "issue_comment" ]; then
+            COMMENT_BODY="$EVENT_COMMENT_BODY"
             ADDITIONAL_INSTRUCTIONS=$(echo "$COMMENT_BODY" | sed 's/.*@gemini//' | xargs)
           fi
           echo "additional_instructions=$ADDITIONAL_INSTRUCTIONS" >> "$GITHUB_OUTPUT"

docker/Dockerfile

@@ -16,7 +16,7 @@ RUN mkdir -p /root/.praison
 # Install Python packages (using latest versions)
 RUN pip install --no-cache-dir \
     flask \
-    "praisonai>=2.3.0" \
+    "praisonai>=2.3.26" \
     "praisonai[api]" \
     gunicorn \
     markdown

docker/Dockerfile.chat

@@ -16,7 +16,7 @@ RUN mkdir -p /root/.praison
 # Install Python packages (using latest versions)
 RUN pip install --no-cache-dir \
     praisonai_tools \
-    "praisonai>=2.3.0" \
+    "praisonai>=2.3.26" \
     "praisonai[chat]" \
     "embedchain[github,youtube]"
 

docker/Dockerfile.dev

@@ -20,7 +20,7 @@ RUN mkdir -p /root/.praison
 # Install Python packages (using latest versions)
 RUN pip install --no-cache-dir \
     praisonai_tools \
-    "praisonai>=2.3.0" \
+    "praisonai>=2.3.26" \
     "praisonai[ui]" \
     "praisonai[chat]" \
     "praisonai[realtime]" \

docker/Dockerfile.ui

@@ -16,7 +16,7 @@ RUN mkdir -p /root/.praison
 # Install Python packages (using latest versions)
 RUN pip install --no-cache-dir \
     praisonai_tools \
-    "praisonai>=2.3.0" \
+    "praisonai>=2.3.26" \
     "praisonai[ui]" \
     "praisonai[crewai]"
 

docker/docker-compose.dev.yml

@@ -137,22 +137,11 @@ services:
       POSTGRES_USER: praisonai
       POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-praisonai_dev_pass}
     volumes:
-   database:
-     image: postgres:15-alpine
-     environment:
-       POSTGRES_DB: praisonai_dev
-       POSTGRES_USER: praisonai
-       POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-praisonai_dev_pass}
-     volumes:
-       - postgres_data:/var/lib/postgresql/data
-     read_only: true
-     tmpfs:
-       - /tmp
-       - /var/tmp
-     ports:
-       - "5432:5432"
-     networks:
       - postgres_data:/var/lib/postgresql/data
+    read_only: true
+    tmpfs:
+      - /tmp
+      - /var/tmp
     ports:
       - "5432:5432"
     networks:

examples/context/00_agent_fast_context_basic.py

@@ -0,0 +1,89 @@
+"""
+Fast Context with Agent - Basic Example
+
+This is the simplest way to use Fast Context with an Agent.
+Just set fast_context=True and the agent will automatically
+have access to rapid code search capabilities.
+
+Fast Context provides:
+- 10-20x faster code search than traditional methods
+- Parallel execution (8 concurrent searches)
+- Automatic caching for repeated queries
+- Seamless integration with Agent workflows
+"""
+
+from praisonaiagents import Agent
+
+def main():
+    print("=" * 60)
+    print("Agent with Fast Context - Basic Example")
+    print("=" * 60)
+    
+    # Create an agent with Fast Context enabled
+    # This is the simplest configuration
+    agent = Agent(
+        name="CodeAssistant",
+        instructions="You are a helpful code assistant.",
+        fast_context=True,  # Enable Fast Context
+        fast_context_path=".",  # Search in current directory
+        verbose=False
+    )
+    
+    print("\n✓ Agent created with Fast Context enabled")
+    print(f"  - fast_context_enabled: {agent.fast_context_enabled}")
+    print(f"  - fast_context_path: {agent.fast_context_path}")
+    print(f"  - fast_context_model: {agent.fast_context_model}")
+    
+    # Use delegate_to_fast_context to search code
+    print("\n" + "-" * 60)
+    print("Searching for 'class Agent' in the codebase...")
+    print("-" * 60)
+    
+    context = agent.delegate_to_fast_context("class Agent")
+    
+    if context:
+        print(f"\n✓ Found relevant code!")
+        print(f"  Context length: {len(context)} characters")
+        print("\nPreview of found context:")
+        print("-" * 40)
+        # Show first 500 characters
+        preview = context[:500]
+        print(preview)
+        if len(context) > 500:
+            print("...")
+    else:
+        print("\n✗ No relevant code found")
+    
+    # Another search example
+    print("\n" + "-" * 60)
+    print("Searching for 'def chat' in the codebase...")
+    print("-" * 60)
+    
+    context = agent.delegate_to_fast_context("def chat")
+    
+    if context:
+        print(f"\n✓ Found relevant code!")
+        print(f"  Context length: {len(context)} characters")
+    else:
+        print("\n✗ No relevant code found")
+    
+    # Access FastContext directly for more control
+    print("\n" + "-" * 60)
+    print("Direct FastContext access for advanced usage...")
+    print("-" * 60)
+    
+    fc = agent.fast_context
+    if fc:
+        result = fc.search("import")
+        print(f"\n✓ Direct search for 'import':")
+        print(f"  Files found: {result.total_files}")
+        print(f"  Search time: {result.search_time_ms}ms")
+        print(f"  From cache: {result.from_cache}")
+    
+    print("\n" + "=" * 60)
+    print("Fast Context makes code search instant for your agents!")
+    print("=" * 60)
+
+
+if __name__ == "__main__":
+    main()

examples/context/01_basic_fast_context.py

@@ -0,0 +1,84 @@
+"""
+Example 1: Basic Fast Context Usage
+
+This example demonstrates the basic usage of Fast Context for rapid code search.
+Fast Context is 10-20x faster than traditional sequential search because it:
+- Executes up to 8 tool calls in parallel
+- Limits search to 4 turns maximum
+- Uses a lightweight model optimized for search
+"""
+
+from praisonaiagents.context.fast import FastContext
+import time
+
+def main():
+    print("=" * 70)
+    print("Fast Context - Basic Usage Example")
+    print("=" * 70)
+    
+    # Create FastContext instance pointing to the praisonai-agents codebase
+    fc = FastContext(
+        workspace_path="/Users/praison/praisonai-package/src/praisonai-agents"
+    )
+    
+    # Example 1: Simple pattern search
+    print("\n1. Simple Pattern Search")
+    print("-" * 40)
+    
+    start = time.perf_counter()
+    result = fc.search("def chat")
+    elapsed = (time.perf_counter() - start) * 1000
+    
+    print(f"   Query: 'def chat'")
+    print(f"   Files found: {result.total_files}")
+    print(f"   Search time: {result.search_time_ms}ms (total: {elapsed:.0f}ms)")
+    print(f"   Turns used: {result.turns_used}")
+    print(f"   Tool calls: {result.total_tool_calls}")
+    
+    if result.files:
+        print("\n   Top matches:")
+        for f in result.files[:3]:
+            print(f"   - {f.path}")
+            for lr in f.line_ranges[:2]:
+                print(f"     Lines {lr.start}-{lr.end}")
+    
+    # Example 2: Search for class definitions
+    print("\n2. Search for Class Definitions")
+    print("-" * 40)
+    
+    result = fc.search("class Agent")
+    print(f"   Query: 'class Agent'")
+    print(f"   Files found: {result.total_files}")
+    print(f"   Search time: {result.search_time_ms}ms")
+    
+    if result.files:
+        print("\n   Files containing 'class Agent':")
+        for f in result.files[:5]:
+            print(f"   - {f.path} (relevance: {f.relevance_score:.2f})")
+    
+    # Example 3: Search for imports
+    print("\n3. Search for Specific Imports")
+    print("-" * 40)
+    
+    result = fc.search("from typing import")
+    print(f"   Query: 'from typing import'")
+    print(f"   Files found: {result.total_files}")
+    print(f"   Search time: {result.search_time_ms}ms")
+    
+    # Example 4: Get formatted context string
+    print("\n4. Get Formatted Context for Agent")
+    print("-" * 40)
+    
+    context = fc.get_context_for_agent("authentication", max_files=3)
+    print(f"   Context length: {len(context)} characters")
+    print(f"   Preview:")
+    preview = context[:300].replace('\n', '\n   ')
+    print(f"   {preview}...")
+    
+    print("\n" + "=" * 70)
+    print("Fast Context provides rapid, parallel code search!")
+    print("=" * 70)
+
+
+if __name__ == "__main__":
+    main()

examples/context/02_parallel_vs_sequential.py

@@ -0,0 +1,113 @@
+"""
+Example 2: Parallel vs Sequential Search Comparison
+
+This example demonstrates the speed advantage of Fast Context's parallel
+execution compared to traditional sequential search.
+
+Key insight: Fast Context executes up to 8 searches simultaneously,
+making it 5-10x faster for complex queries.
+"""
+
+import time
+import os
+from praisonaiagents.context.fast import FastContext
+from praisonaiagents.context.fast.search_tools import grep_search, glob_search
+
+WORKSPACE = "/Users/praison/praisonai-package/src/praisonai-agents"
+
+
+def sequential_search(queries: list) -> dict:
+    """Traditional sequential search - one query at a time."""
+    results = {}
+    for query in queries:
+        results[query] = grep_search(
+            search_path=WORKSPACE,
+            pattern=query,
+            max_results=20
+        )
+    return results
+
+
+def parallel_search_with_fast_context(queries: list) -> dict:
+    """Fast Context parallel search - multiple queries at once."""
+    fc = FastContext(workspace_path=WORKSPACE)
+    results = {}
+    for query in queries:
+        result = fc.search(query)
+        results[query] = result
+    return results
+
+
+def main():
+    print("=" * 70)
+    print("Parallel vs Sequential Search Comparison")
+    print("=" * 70)
+    
+    # Define multiple search queries
+    queries = [
+        "def __init__",
+        "class Agent",
+        "async def",
+        "import logging",
+        "from typing",
+    ]
+    
+    print(f"\nSearching for {len(queries)} patterns in the codebase...")
+    print(f"Workspace: {WORKSPACE}")
+    
+    # Sequential search
+    print("\n1. Sequential Search (Traditional)")
+    print("-" * 40)
+    
+    start = time.perf_counter()
+    seq_results = sequential_search(queries)
+    seq_time = (time.perf_counter() - start) * 1000
+    
+    total_seq_matches = sum(len(r) for r in seq_results.values())
+    print(f"   Total matches: {total_seq_matches}")
+    print(f"   Time: {seq_time:.0f}ms")
+    
+    # Fast Context search (with caching disabled for fair comparison)
+    print("\n2. Fast Context Search (Parallel)")
+    print("-" * 40)
+    
+    fc = FastContext(workspace_path=WORKSPACE, cache_enabled=False)
+    
+    start = time.perf_counter()
+    fc_results = {}
+    for query in queries:
+        fc_results[query] = fc.search(query)
+    fc_time = (time.perf_counter() - start) * 1000
+    
+    total_fc_matches = sum(r.total_files for r in fc_results.values())
+    print(f"   Total files found: {total_fc_matches}")
+    print(f"   Time: {fc_time:.0f}ms")
+    
+    # Comparison
+    print("\n3. Comparison")
+    print("-" * 40)
+    
+    speedup = seq_time / fc_time if fc_time > 0 else float('inf')
+    print(f"   Sequential: {seq_time:.0f}ms")
+    print(f"   Fast Context: {fc_time:.0f}ms")
+    print(f"   Speedup: {speedup:.1f}x faster")
+    
+    # Detailed results
+    print("\n4. Results by Query")
+    print("-" * 40)
+    
+    for query in queries:
+        seq_count = len(seq_results[query])
+        fc_count = fc_results[query].total_files
+        fc_ms = fc_results[query].search_time_ms
+        print(f"   '{query}':")
+        print(f"      Sequential: {seq_count} matches")
+        print(f"      Fast Context: {fc_count} files in {fc_ms}ms")
+    
+    print("\n" + "=" * 70)
+    print(f"Fast Context is {speedup:.1f}x faster due to parallel execution!")
+    print("=" * 70)
+
+
+if __name__ == "__main__":
+    main()

examples/context/03_caching_performance.py

@@ -0,0 +1,132 @@
+"""
+Example 3: Caching Performance Demonstration
+
+This example shows how Fast Context's caching dramatically improves
+performance for repeated queries.
+
+Key benefits:
+- First query: Full search execution
+- Subsequent queries: Instant cache hit (0ms search time)
+- Configurable TTL (time-to-live)
+"""
+
+import time
+from praisonaiagents.context.fast import FastContext
+
+WORKSPACE = "/Users/praison/praisonai-package/src/praisonai-agents"
+
+
+def main():
+    print("=" * 70)
+    print("Fast Context Caching Performance")
+    print("=" * 70)
+    
+    # Create FastContext with caching enabled
+    fc = FastContext(
+        workspace_path=WORKSPACE,
+        cache_enabled=True,
+        cache_ttl=300  # 5 minutes
+    )
+    
+    query = "def execute"
+    
+    print(f"\nQuery: '{query}'")
+    print(f"Cache TTL: 300 seconds")
+    
+    # First search - cache miss
+    print("\n1. First Search (Cache Miss)")
+    print("-" * 40)
+    
+    start = time.perf_counter()
+    result1 = fc.search(query)
+    elapsed1 = (time.perf_counter() - start) * 1000
+    
+    print(f"   Files found: {result1.total_files}")
+    print(f"   From cache: {result1.from_cache}")
+    print(f"   Search time: {result1.search_time_ms}ms")
+    print(f"   Total time: {elapsed1:.1f}ms")
+    
+    # Second search - cache hit
+    print("\n2. Second Search (Cache Hit)")
+    print("-" * 40)
+    
+    start = time.perf_counter()
+    result2 = fc.search(query)
+    elapsed2 = (time.perf_counter() - start) * 1000
+    
+    print(f"   Files found: {result2.total_files}")
+    print(f"   From cache: {result2.from_cache}")
+    print(f"   Search time: {result2.search_time_ms}ms")
+    print(f"   Total time: {elapsed2:.1f}ms")
+    
+    # Third search - still cached
+    print("\n3. Third Search (Still Cached)")
+    print("-" * 40)
+    
+    start = time.perf_counter()
+    result3 = fc.search(query)
+    elapsed3 = (time.perf_counter() - start) * 1000
+    
+    print(f"   Files found: {result3.total_files}")
+    print(f"   From cache: {result3.from_cache}")
+    print(f"   Total time: {elapsed3:.1f}ms")
+    
+    # Clear cache and search again
+    print("\n4. After Cache Clear")
+    print("-" * 40)
+    
+    fc.clear_cache()
+    print("   Cache cleared!")
+    
+    start = time.perf_counter()
+    result4 = fc.search(query)
+    elapsed4 = (time.perf_counter() - start) * 1000
+    
+    print(f"   Files found: {result4.total_files}")
+    print(f"   From cache: {result4.from_cache}")
+    print(f"   Search time: {result4.search_time_ms}ms")
+    print(f"   Total time: {elapsed4:.1f}ms")
+    
+    # Performance comparison
+    print("\n5. Performance Summary")
+    print("-" * 40)
+    
+    cache_speedup = elapsed1 / elapsed2 if elapsed2 > 0 else float('inf')
+    
+    print(f"   First search (no cache): {elapsed1:.1f}ms")
+    print(f"   Cached searches: ~{elapsed2:.1f}ms")
+    print(f"   Cache speedup: {cache_speedup:.0f}x faster")
+    
+    # Multiple different queries
+    print("\n6. Multiple Query Caching")
+    print("-" * 40)
+    
+    queries = ["class Agent", "async def", "import os", "def __init__"]
+    
+    # First pass - populate cache
+    print("   First pass (populating cache):")
+    start = time.perf_counter()
+    for q in queries:
+        fc.search(q)
+    first_pass = (time.perf_counter() - start) * 1000
+    print(f"   Time: {first_pass:.0f}ms")
+    
+    # Second pass - all cached
+    print("   Second pass (all cached):")
+    start = time.perf_counter()
+    for q in queries:
+        r = fc.search(q)
+        assert r.from_cache, f"Expected cache hit for '{q}'"
+    second_pass = (time.perf_counter() - start) * 1000
+    print(f"   Time: {second_pass:.1f}ms")
+    
+    multi_speedup = first_pass / second_pass if second_pass > 0 else float('inf')
+    print(f"   Speedup: {multi_speedup:.0f}x faster")
+    
+    print("\n" + "=" * 70)
+    print("Caching provides instant results for repeated queries!")
+    print("=" * 70)
+
+
+if __name__ == "__main__":
+    main()

examples/context/04_agent_integration.py

@@ -0,0 +1,112 @@
+"""
+Example 4: Agent Integration with Fast Context
+
+This example shows how to use Fast Context integrated with the Agent class
+for automatic code search delegation.
+
+Benefits:
+- Seamless integration with existing Agent workflows
+- Automatic context injection for code-related queries
+- Configurable via Agent parameters or environment variables
+"""
+
+from praisonaiagents import Agent
+
+WORKSPACE = "/Users/praison/praisonai-package/src/praisonai-agents"
+
+
+def main():
+    print("=" * 70)
+    print("Agent Integration with Fast Context")
+    print("=" * 70)
+    
+    # Create an agent with Fast Context enabled
+    print("\n1. Creating Agent with Fast Context")
+    print("-" * 40)
+    
+    agent = Agent(
+        name="CodeAssistant",
+        instructions="You are a helpful code assistant that can search and analyze code.",
+        fast_context=True,
+        fast_context_path=WORKSPACE,
+        fast_context_model="gpt-4o-mini",
+        fast_context_max_turns=4,
+        fast_context_parallelism=8,
+        fast_context_timeout=30.0,
+        verbose=False
+    )
+    
+    print(f"   Agent name: {agent.name}")
+    print(f"   Fast Context enabled: {agent.fast_context_enabled}")
+    print(f"   Fast Context path: {agent.fast_context_path}")
+    print(f"   Fast Context model: {agent.fast_context_model}")
+    print(f"   Max turns: {agent.fast_context_max_turns}")
+    print(f"   Parallelism: {agent.fast_context_parallelism}")
+    
+    # Use delegate_to_fast_context for code search
+    print("\n2. Delegating Code Search to Fast Context")
+    print("-" * 40)
+    
+    queries = [
+        "find the Agent class definition",
+        "search for authentication handlers",
+        "locate memory management code",
+    ]
+    
+    for query in queries:
+        print(f"\n   Query: '{query}'")
+        context = agent.delegate_to_fast_context(query)
+        
+        if context:
+            # Show preview of context
+            lines = context.split('\n')
+            preview_lines = lines[:5]
+            print(f"   Found context ({len(context)} chars):")
+            for line in preview_lines:
+                print(f"      {line[:60]}{'...' if len(line) > 60 else ''}")
+            if len(lines) > 5:
+                print(f"      ... and {len(lines) - 5} more lines")
+        else:
+            print("   No relevant code found")
+    
+    # Access the FastContext instance directly
+    print("\n3. Direct FastContext Access")
+    print("-" * 40)
+    
+    fc = agent.fast_context
+    if fc:
+        result = fc.search("def chat")
+        print(f"   Direct search for 'def chat':")
+        print(f"   Files found: {result.total_files}")
+        print(f"   Search time: {result.search_time_ms}ms")
+        
+        for f in result.files[:3]:
+            print(f"   - {f.path}")
+    
+    # Show configuration options
+    print("\n4. Configuration Options")
+    print("-" * 40)
+    
+    print("   Agent parameters:")
+    print("   - fast_context=True              # Enable Fast Context")
+    print("   - fast_context_path='.'          # Workspace path")
+    print("   - fast_context_model='gpt-4o-mini'  # Search model")
+    print("   - fast_context_max_turns=4       # Max search turns")
+    print("   - fast_context_parallelism=8     # Parallel calls")
+    print("   - fast_context_timeout=30.0      # Timeout (seconds)")
+    
+    print("\n   Environment variables:")
+    print("   - FAST_CONTEXT_MODEL")
+    print("   - FAST_CONTEXT_MAX_TURNS")
+    print("   - FAST_CONTEXT_PARALLELISM")
+    print("   - FAST_CONTEXT_TIMEOUT")
+    print("   - FAST_CONTEXT_CACHE")
+    print("   - FAST_CONTEXT_CACHE_TTL")
+    
+    print("\n" + "=" * 70)
+    print("Fast Context integrates seamlessly with Agent for code search!")
+    print("=" * 70)
+
+
+if __name__ == "__main__":
+    main()

examples/context/05_file_and_symbol_indexing.py

@@ -0,0 +1,121 @@
+"""
+Example 5: File and Symbol Indexing
+
+This example demonstrates the file and symbol indexing capabilities
+that power Fast Context's rapid search.
+
+Features:
+- FileIndexer: Index file metadata for fast lookups
+- SymbolIndexer: Extract code symbols (functions, classes, imports)
+- Support for Python, JavaScript, TypeScript, Go, Rust, Java
+"""
+
+import time
+from praisonaiagents.context.fast.indexer import (
+    FileIndexer,
+    SymbolIndexer,
+    SymbolType
+)
+
+WORKSPACE = "/Users/praison/praisonai-package/src/praisonai-agents"
+
+
+def main():
+    print("=" * 70)
+    print("File and Symbol Indexing")
+    print("=" * 70)
+    
+    # File Indexer
+    print("\n1. File Indexer")
+    print("-" * 40)
+    
+    start = time.perf_counter()
+    file_indexer = FileIndexer(workspace_path=WORKSPACE)
+    file_count = file_indexer.index()
+    file_time = (time.perf_counter() - start) * 1000
+    
+    print(f"   Indexed {file_count} files in {file_time:.0f}ms")
+    
+    stats = file_indexer.get_stats()
+    print(f"   Total size: {stats['total_size_mb']} MB")
+    print(f"   Extensions: {list(stats['extensions'].keys())[:5]}")
+    
+    # Find files by pattern
+    print("\n   Finding files by pattern:")
+    py_files = file_indexer.find_by_pattern("**/*.py")
+    print(f"   - Python files (*.py): {len(py_files)}")
+    
+    agent_files = file_indexer.find_by_pattern("**/agent*.py")
+    print(f"   - Agent files (agent*.py): {len(agent_files)}")
+    
+    # Find files by name
+    print("\n   Finding files by name:")
+    init_files = file_indexer.find_by_name("__init__.py", exact=True)
+    print(f"   - __init__.py files: {len(init_files)}")
+    
+    agent_matches = file_indexer.find_by_name("agent")
+    print(f"   - Files containing 'agent': {len(agent_matches)}")
+    
+    # Symbol Indexer
+    print("\n2. Symbol Indexer")
+    print("-" * 40)
+    
+    start = time.perf_counter()
+    symbol_indexer = SymbolIndexer(workspace_path=WORKSPACE)
+    symbol_count = symbol_indexer.index()
+    symbol_time = (time.perf_counter() - start) * 1000
+    
+    print(f"   Indexed {symbol_count} symbols in {symbol_time:.0f}ms")
+    
+    stats = symbol_indexer.get_stats()
+    print(f"   Files indexed: {stats['total_files']}")
+    print(f"   Symbols by type:")
+    for sym_type, count in stats['by_type'].items():
+        print(f"      - {sym_type}: {count}")
+    
+    # Find symbols by name
+    print("\n   Finding symbols by name:")
+    
+    agent_symbols = symbol_indexer.find_by_name("Agent", exact=True)
+    print(f"   - 'Agent' (exact): {len(agent_symbols)} symbols")
+    for sym in agent_symbols[:3]:
+        print(f"      {sym.symbol_type.value}: {sym.name} in {sym.file_path}:{sym.line_number}")
+    
+    chat_symbols = symbol_indexer.find_by_name("chat")
+    print(f"   - 'chat' (partial): {len(chat_symbols)} symbols")
+    for sym in chat_symbols[:3]:
+        print(f"      {sym.symbol_type.value}: {sym.name} in {sym.file_path}:{sym.line_number}")
+    
+    # Find symbols by type
+    print("\n   Finding symbols by type:")
+    
+    classes = symbol_indexer.find_by_type(SymbolType.CLASS)
+    print(f"   - Classes: {len(classes)}")
+    for cls in classes[:5]:
+        print(f"      class {cls.name} in {cls.file_path}:{cls.line_number}")
+    
+    functions = symbol_indexer.find_by_type(SymbolType.FUNCTION)
+    print(f"   - Functions: {len(functions)}")
+    
+    methods = symbol_indexer.find_by_type(SymbolType.METHOD)
+    print(f"   - Methods: {len(methods)}")
+    
+    imports = symbol_indexer.find_by_type(SymbolType.IMPORT)
+    print(f"   - Imports: {len(imports)}")
+    
+    # Performance comparison
+    print("\n3. Performance Summary")
+    print("-" * 40)
+    
+    print(f"   File indexing: {file_time:.0f}ms for {file_count} files")
+    print(f"   Symbol indexing: {symbol_time:.0f}ms for {symbol_count} symbols")
+    print(f"   Files per second: {file_count / (file_time / 1000):.0f}")
+    print(f"   Symbols per second: {symbol_count / (symbol_time / 1000):.0f}")
+    
+    print("\n" + "=" * 70)
+    print("Indexing enables instant lookups for files and symbols!")
+    print("=" * 70)
+
+
+if __name__ == "__main__":
+    main()

examples/context/06_context_injection.py

@@ -0,0 +1,163 @@
+"""
+Example 6: Context Injection for Agents
+
+This example demonstrates how Fast Context results can be formatted
+and injected into agent prompts for better code understanding.
+
+Features:
+- Multiple format styles (markdown, xml, plain)
+- Token budget management
+- Precision over recall (avoid context pollution)
+"""
+
+from praisonaiagents.context.fast import FastContext
+from praisonaiagents.context.fast.context_injector import (
+    ContextInjector,
+    InjectionConfig,
+    inject_fast_context
+)
+
+WORKSPACE = "/Users/praison/praisonai-package/src/praisonai-agents"
+
+
+def main():
+    print("=" * 70)
+    print("Context Injection for Agents")
+    print("=" * 70)
+    
+    # Search for code
+    fc = FastContext(workspace_path=WORKSPACE)
+    result = fc.search("class Agent")
+    
+    print(f"\nSearch results: {result.total_files} files found")
+    
+    # Markdown format (default)
+    print("\n1. Markdown Format")
+    print("-" * 40)
+    
+    config_md = InjectionConfig(
+        max_tokens=2000,
+        max_files=3,
+        max_lines_per_file=20,
+        format_style="markdown",
+        include_line_numbers=True,
+        include_file_content=True,
+        prioritize_precision=True
+    )
+    
+    injector_md = ContextInjector(config_md)
+    md_context = injector_md.format_context(result)
+    
+    print(f"   Format: Markdown")
+    print(f"   Length: {len(md_context)} chars (~{len(md_context)//4} tokens)")
+    print(f"   Preview:")
+    for line in md_context.split('\n')[:10]:
+        print(f"   {line[:70]}")
+    print("   ...")
+    
+    # XML format
+    print("\n2. XML Format")
+    print("-" * 40)
+    
+    config_xml = InjectionConfig(
+        max_tokens=2000,
+        max_files=3,
+        format_style="xml",
+        prioritize_precision=True
+    )
+    
+    injector_xml = ContextInjector(config_xml)
+    xml_context = injector_xml.format_context(result)
+    
+    print(f"   Format: XML")
+    print(f"   Length: {len(xml_context)} chars")
+    print(f"   Preview:")
+    for line in xml_context.split('\n')[:10]:
+        print(f"   {line[:70]}")
+    print("   ...")
+    
+    # Plain text format
+    print("\n3. Plain Text Format")
+    print("-" * 40)
+    
+    config_plain = InjectionConfig(
+        max_tokens=2000,
+        max_files=3,
+        format_style="plain"
+    )
+    
+    injector_plain = ContextInjector(config_plain)
+    plain_context = injector_plain.format_context(result)
+    
+    print(f"   Format: Plain")
+    print(f"   Length: {len(plain_context)} chars")
+    print(f"   Preview:")
+    for line in plain_context.split('\n')[:8]:
+        print(f"   {line[:70]}")
+    print("   ...")
+    
+    # Inject into prompts
+    print("\n4. Injecting into Prompts")
+    print("-" * 40)
+    
+    system_prompt = "You are a helpful code assistant."
+    user_message = "Explain how the Agent class works."
+    
+    injected = inject_fast_context(
+        result=result,
+        system_prompt=system_prompt,
+        user_message=user_message,
+        max_tokens=1500,
+        format_style="markdown"
+    )
+    
+    print(f"   Original system prompt: {len(system_prompt)} chars")
+    print(f"   Injected system prompt: {len(injected['system_prompt'])} chars")
+    print(f"   User message unchanged: {injected['user_message'] == user_message}")
+    
+    # Token budget management
+    print("\n5. Token Budget Management")
+    print("-" * 40)
+    
+    budgets = [500, 1000, 2000, 4000]
+    
+    for budget in budgets:
+        config = InjectionConfig(max_tokens=budget, max_files=10)
+        injector = ContextInjector(config)
+        context = injector.format_context(result)
+        estimated_tokens = len(context) // 4
+        print(f"   Budget: {budget} tokens -> {len(context)} chars (~{estimated_tokens} tokens)")
+    
+    # Precision mode
+    print("\n6. Precision vs Recall")
+    print("-" * 40)
+    
+    # High precision (fewer, more relevant results)
+    config_precision = InjectionConfig(
+        max_tokens=2000,
+        max_files=5,
+        prioritize_precision=True
+    )
+    injector_precision = ContextInjector(config_precision)
+    precision_context = injector_precision.format_context(result)
+    
+    # Lower precision (more results)
+    config_recall = InjectionConfig(
+        max_tokens=2000,
+        max_files=10,
+        prioritize_precision=False
+    )
+    injector_recall = ContextInjector(config_recall)
+    recall_context = injector_recall.format_context(result)
+    
+    print(f"   High precision mode: {len(precision_context)} chars")
+    print(f"   High recall mode: {len(recall_context)} chars")
+    print(f"   Precision mode is more focused for better agent understanding")
+    
+    print("\n" + "=" * 70)
+    print("Context injection formats search results for optimal agent use!")
+    print("=" * 70)
+
+
+if __name__ == "__main__":
+    main()

examples/context/07_search_tools_demo.py

@@ -0,0 +1,213 @@
+"""
+Example 7: Search Tools Deep Dive
+
+This example demonstrates the individual search tools that power Fast Context:
+- grep_search: Pattern-based search with regex support
+- glob_search: File pattern matching
+- read_file: Read file contents with line ranges
+- list_directory: Directory listing with filtering
+"""
+
+import time
+from praisonaiagents.context.fast.search_tools import (
+    grep_search,
+    glob_search,
+    read_file,
+    list_directory
+)
+
+WORKSPACE = "/Users/praison/praisonai-package/src/praisonai-agents"
+
+
+def main():
+    print("=" * 70)
+    print("Search Tools Deep Dive")
+    print("=" * 70)
+    
+    # grep_search
+    print("\n1. grep_search - Pattern Search")
+    print("-" * 40)
+    
+    # Simple pattern
+    start = time.perf_counter()
+    results = grep_search(
+        search_path=WORKSPACE,
+        pattern="def __init__",
+        max_results=10
+    )
+    elapsed = (time.perf_counter() - start) * 1000
+    
+    print(f"   Pattern: 'def __init__'")
+    print(f"   Matches: {len(results)}")
+    print(f"   Time: {elapsed:.0f}ms")
+    if results:
+        print(f"   First match: {results[0]['path']}:{results[0]['line_number']}")
+    
+    # Regex pattern
+    results = grep_search(
+        search_path=WORKSPACE,
+        pattern=r"class \w+Agent",
+        is_regex=True,
+        max_results=10
+    )
+    print(f"\n   Regex: 'class \\w+Agent'")
+    print(f"   Matches: {len(results)}")
+    for r in results[:3]:
+        print(f"   - {r['path']}:{r['line_number']}: {r['content'][:50]}...")
+    
+    # Case sensitive
+    results_sensitive = grep_search(
+        search_path=WORKSPACE,
+        pattern="Agent",
+        case_sensitive=True,
+        max_results=50
+    )
+    results_insensitive = grep_search(
+        search_path=WORKSPACE,
+        pattern="agent",
+        case_sensitive=False,
+        max_results=50
+    )
+    print(f"\n   Case sensitive 'Agent': {len(results_sensitive)} matches")
+    print(f"   Case insensitive 'agent': {len(results_insensitive)} matches")
+    
+    # With context lines
+    results = grep_search(
+        search_path=WORKSPACE,
+        pattern="class Agent:",
+        context_lines=2,
+        max_results=3
+    )
+    print(f"\n   With 2 context lines:")
+    if results:
+        print(f"   {results[0]['path']}:")
+        if results[0].get('context'):
+            for line in results[0]['context'].split('\n')[:5]:
+                print(f"      {line[:60]}")
+    
+    # glob_search
+    print("\n2. glob_search - File Pattern Matching")
+    print("-" * 40)
+    
+    # Find Python files
+    start = time.perf_counter()
+    results = glob_search(
+        search_path=WORKSPACE,
+        pattern="**/*.py",
+        max_results=100
+    )
+    elapsed = (time.perf_counter() - start) * 1000
+    
+    print(f"   Pattern: '**/*.py'")
+    print(f"   Files found: {len(results)}")
+    print(f"   Time: {elapsed:.0f}ms")
+    
+    # Find specific files
+    results = glob_search(
+        search_path=WORKSPACE,
+        pattern="**/agent*.py",
+        max_results=20
+    )
+    print(f"\n   Pattern: '**/agent*.py'")
+    print(f"   Files found: {len(results)}")
+    for r in results[:5]:
+        print(f"   - {r['path']} ({r['size']} bytes)")
+    
+    # Find test files
+    results = glob_search(
+        search_path=WORKSPACE,
+        pattern="**/test_*.py",
+        max_results=50
+    )
+    print(f"\n   Pattern: '**/test_*.py'")
+    print(f"   Test files: {len(results)}")
+    
+    # read_file
+    print("\n3. read_file - Read File Contents")
+    print("-" * 40)
+    
+    # Read entire file
+    result = read_file(
+        filepath=f"{WORKSPACE}/praisonaiagents/__init__.py",
+        max_lines=20
+    )
+    
+    print(f"   File: praisonaiagents/__init__.py")
+    print(f"   Success: {result['success']}")
+    print(f"   Lines: {result.get('total_lines', 'N/A')}")
+    if result['success']:
+        lines = result['content'].split('\n')[:5]
+        print(f"   Preview:")
+        for line in lines:
+            print(f"      {line[:60]}")
+    
+    # Read specific line range
+    result = read_file(
+        filepath=f"{WORKSPACE}/praisonaiagents/agent/agent.py",
+        start_line=1,
+        end_line=20
+    )
+    
+    print(f"\n   File: agent/agent.py (lines 1-20)")
+    print(f"   Success: {result['success']}")
+    if result['success']:
+        lines = result['content'].split('\n')[:5]
+        print(f"   Preview:")
+        for line in lines:
+            print(f"      {line[:60]}")
+    
+    # Read with context
+    result = read_file(
+        filepath=f"{WORKSPACE}/praisonaiagents/agent/agent.py",
+        start_line=100,
+        end_line=105,
+        context_lines=3
+    )
+    
+    print(f"\n   File: agent/agent.py (lines 100-105 with 3 context lines)")
+    print(f"   Success: {result['success']}")
+    
+    # list_directory
+    print("\n4. list_directory - Directory Listing")
+    print("-" * 40)
+    
+    # List root
+    result = list_directory(
+        dir_path=WORKSPACE,
+        recursive=False,
+        max_entries=20
+    )
+    
+    print(f"   Directory: {WORKSPACE}")
+    print(f"   Success: {result['success']}")
+    print(f"   Entries: {len(result.get('entries', []))}")
+    
+    dirs = [e for e in result.get('entries', []) if e.get('is_dir')]
+    files = [e for e in result.get('entries', []) if not e.get('is_dir')]
+    print(f"   Directories: {len(dirs)}")
+    print(f"   Files: {len(files)}")
+    
+    # List recursive with depth limit
+    result = list_directory(
+        dir_path=f"{WORKSPACE}/praisonaiagents",
+        recursive=True,
+        max_depth=2,
+        max_entries=50
+    )
+    
+    print(f"\n   Directory: praisonaiagents (recursive, depth=2)")
+    print(f"   Entries: {len(result.get('entries', []))}")
+    
+    # Show directory structure
+    print(f"   Structure:")
+    for entry in result.get('entries', [])[:10]:
+        prefix = "📁" if entry.get('is_dir') else "📄"
+        print(f"      {prefix} {entry['name']}")
+    
+    print("\n" + "=" * 70)
+    print("These search tools power Fast Context's rapid code search!")
+    print("=" * 70)
+
+
+if __name__ == "__main__":
+    main()

examples/context/08_comprehensive_benchmark.py

@@ -0,0 +1,227 @@
+"""
+Example 8: Comprehensive Benchmark
+
+This example provides a comprehensive benchmark comparing Fast Context
+performance against traditional search approaches.
+
+Metrics measured:
+- Search latency
+- Throughput (queries per second)
+- Cache effectiveness
+- Parallel execution speedup
+"""
+
+import time
+import statistics
+from praisonaiagents.context.fast import FastContext
+from praisonaiagents.context.fast.search_tools import grep_search
+from praisonaiagents.context.fast.indexer import FileIndexer, SymbolIndexer
+
+WORKSPACE = "/Users/praison/praisonai-package/src/praisonai-agents"
+
+
+def benchmark_search_latency():
+    """Benchmark search latency."""
+    print("\n1. Search Latency Benchmark")
+    print("-" * 50)
+    
+    fc = FastContext(workspace_path=WORKSPACE, cache_enabled=False)
+    
+    queries = [
+        "def __init__",
+        "class Agent",
+        "import logging",
+        "async def",
+        "from typing import",
+    ]
+    
+    # Warm up
+    fc.search("warmup")
+    
+    # Benchmark each query
+    latencies = []
+    for query in queries:
+        times = []
+        for _ in range(3):  # 3 runs per query
+            start = time.perf_counter()
+            fc.search(query)
+            elapsed = (time.perf_counter() - start) * 1000
+            times.append(elapsed)
+        
+        avg = statistics.mean(times)
+        latencies.append(avg)
+        print(f"   '{query}': {avg:.1f}ms (min: {min(times):.1f}, max: {max(times):.1f})")
+    
+    print(f"\n   Average latency: {statistics.mean(latencies):.1f}ms")
+    print(f"   Median latency: {statistics.median(latencies):.1f}ms")
+    print(f"   P95 latency: {sorted(latencies)[int(len(latencies)*0.95)]:.1f}ms")
+
+
+def benchmark_throughput():
+    """Benchmark queries per second."""
+    print("\n2. Throughput Benchmark")
+    print("-" * 50)
+    
+    fc = FastContext(workspace_path=WORKSPACE, cache_enabled=True)
+    
+    queries = ["def", "class", "import", "return", "if", "for", "while", "try"]
+    
+    # Cold run (no cache)
+    fc.clear_cache()
+    start = time.perf_counter()
+    for query in queries:
+        fc.search(query)
+    cold_time = time.perf_counter() - start
+    cold_qps = len(queries) / cold_time
+    
+    print(f"   Cold (no cache):")
+    print(f"      Queries: {len(queries)}")
+    print(f"      Time: {cold_time*1000:.0f}ms")
+    print(f"      Throughput: {cold_qps:.1f} queries/second")
+    
+    # Warm run (cached)
+    start = time.perf_counter()
+    for _ in range(10):  # 10 iterations
+        for query in queries:
+            fc.search(query)
+    warm_time = time.perf_counter() - start
+    warm_qps = (len(queries) * 10) / warm_time
+    
+    print(f"\n   Warm (cached):")
+    print(f"      Queries: {len(queries) * 10}")
+    print(f"      Time: {warm_time*1000:.0f}ms")
+    print(f"      Throughput: {warm_qps:.0f} queries/second")
+    
+    print(f"\n   Cache speedup: {warm_qps/cold_qps:.0f}x")
+
+
+def benchmark_cache_effectiveness():
+    """Benchmark cache hit rate and speedup."""
+    print("\n3. Cache Effectiveness Benchmark")
+    print("-" * 50)
+    
+    fc = FastContext(workspace_path=WORKSPACE, cache_enabled=True)
+    fc.clear_cache()
+    
+    queries = ["Agent", "chat", "memory", "tool", "execute"]
+    
+    # First pass - cache misses
+    miss_times = []
+    for query in queries:
+        start = time.perf_counter()
+        result = fc.search(query)
+        elapsed = (time.perf_counter() - start) * 1000
+        miss_times.append(elapsed)
+        assert not result.from_cache
+    
+    # Second pass - cache hits
+    hit_times = []
+    for query in queries:
+        start = time.perf_counter()
+        result = fc.search(query)
+        elapsed = (time.perf_counter() - start) * 1000
+        hit_times.append(elapsed)
+        assert result.from_cache
+    
+    avg_miss = statistics.mean(miss_times)
+    avg_hit = statistics.mean(hit_times)
+    
+    print(f"   Cache miss (first query): {avg_miss:.1f}ms avg")
+    print(f"   Cache hit (repeat query): {avg_hit:.2f}ms avg")
+    print(f"   Cache speedup: {avg_miss/avg_hit:.0f}x faster")
+    print(f"   Time saved per cached query: {avg_miss - avg_hit:.1f}ms")
+
+
+def benchmark_parallel_execution():
+    """Benchmark parallel vs sequential execution."""
+    print("\n4. Parallel Execution Benchmark")
+    print("-" * 50)
+    
+    queries = ["def", "class", "import", "return", "async"]
+    
+    # Sequential (one at a time)
+    start = time.perf_counter()
+    for query in queries:
+        grep_search(WORKSPACE, query, max_results=20)
+    seq_time = (time.perf_counter() - start) * 1000
+    
+    # Parallel (Fast Context)
+    fc = FastContext(workspace_path=WORKSPACE, cache_enabled=False)
+    start = time.perf_counter()
+    for query in queries:
+        fc.search(query)
+    par_time = (time.perf_counter() - start) * 1000
+    
+    print(f"   Sequential: {seq_time:.0f}ms")
+    print(f"   Parallel (Fast Context): {par_time:.0f}ms")
+    print(f"   Speedup: {seq_time/par_time:.1f}x")
+
+
+def benchmark_indexing():
+    """Benchmark indexing performance."""
+    print("\n5. Indexing Performance Benchmark")
+    print("-" * 50)
+    
+    # File indexing
+    start = time.perf_counter()
+    file_indexer = FileIndexer(workspace_path=WORKSPACE)
+    file_count = file_indexer.index()
+    file_time = (time.perf_counter() - start) * 1000
+    
+    print(f"   File Indexer:")
+    print(f"      Files: {file_count}")
+    print(f"      Time: {file_time:.0f}ms")
+    print(f"      Rate: {file_count/(file_time/1000):.0f} files/second")
+    
+    # Symbol indexing
+    start = time.perf_counter()
+    symbol_indexer = SymbolIndexer(workspace_path=WORKSPACE)
+    symbol_count = symbol_indexer.index()
+    symbol_time = (time.perf_counter() - start) * 1000
+    
+    print(f"\n   Symbol Indexer:")
+    print(f"      Symbols: {symbol_count}")
+    print(f"      Time: {symbol_time:.0f}ms")
+    print(f"      Rate: {symbol_count/(symbol_time/1000):.0f} symbols/second")
+    
+    # Lookup performance
+    start = time.perf_counter()
+    for _ in range(100):
+        file_indexer.find_by_pattern("**/*.py")
+    lookup_time = (time.perf_counter() - start) * 1000
+    
+    print(f"\n   Lookup Performance:")
+    print(f"      100 pattern lookups: {lookup_time:.1f}ms")
+    print(f"      Per lookup: {lookup_time/100:.2f}ms")
+
+
+def main():
+    print("=" * 70)
+    print("Fast Context Comprehensive Benchmark")
+    print("=" * 70)
+    print(f"\nWorkspace: {WORKSPACE}")
+    
+    benchmark_search_latency()
+    benchmark_throughput()
+    benchmark_cache_effectiveness()
+    benchmark_parallel_execution()
+    benchmark_indexing()
+    
+    print("\n" + "=" * 70)
+    print("Summary")
+    print("=" * 70)
+    print("""
+   Fast Context provides:
+   ✓ Sub-100ms search latency for most queries
+   ✓ 100+ queries/second with caching
+   ✓ 100x+ speedup for cached queries
+   ✓ 2-5x speedup from parallel execution
+   ✓ 1000+ files/second indexing rate
+   
+   These benchmarks demonstrate why Fast Context is ideal for
+   rapid code search in AI agent workflows.
+""")
+
+
+if __name__ == "__main__":
+    main()

examples/cookbooks/Finance_Market_Job_Agents/Ai_Job_Hunting_Assistant.ipynb

@@ -29,7 +29,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 1,
+   "execution_count": null,
    "metadata": {
     "colab": {
      "base_uri": "https://localhost:8080/"
@@ -71,7 +71,6 @@
     "import pandas as pd\n",
     "\n",
     "print(\"📦 Dependencies installed!\")\n",
-    "print(\"🔄 Converting from AGNO → PraisonAI\")\n",
     "print(\"🎯 Building Job Hunting Agent System\")"
    ]
   },

examples/python/mcp/mcp-backward-compat.py

@@ -0,0 +1,83 @@
+"""
+MCP Backward Compatibility Example
+
+Demonstrates backward compatibility with older MCP servers
+using the deprecated HTTP+SSE transport (protocol version 2024-11-05).
+
+Features:
+- Legacy SSE endpoint detection (/sse suffix)
+- Automatic transport negotiation
+- Protocol version support (2024-11-05 to 2025-11-25)
+
+Protocol: MCP 2025-11-25 with 2024-11-05 compatibility
+"""
+
+# Example code (requires running MCP servers):
+#
+# from praisonaiagents import Agent, MCP
+#
+# # Legacy SSE server (auto-detected via /sse suffix)
+# agent_legacy = Agent(
+#     name="Legacy SSE Assistant",
+#     tools=MCP("http://localhost:8080/sse")
+# )
+#
+# # Modern Streamable HTTP server
+# agent_modern = Agent(
+#     name="Modern HTTP Assistant",
+#     tools=MCP("http://localhost:8080/mcp")
+# )
+
+if __name__ == "__main__":
+    from praisonaiagents.mcp.mcp_transport import get_transport_type
+    from praisonaiagents.mcp.mcp_compat import (
+        detect_transport_support,
+        is_legacy_sse_url,
+        TransportNegotiator
+    )
+    from praisonaiagents.mcp.mcp_session import VALID_PROTOCOL_VERSIONS
+    
+    print("MCP Backward Compatibility Example")
+    print("=" * 50)
+    
+    # URL detection
+    print("\n1. URL-based Transport Detection:")
+    urls = [
+        "http://localhost:8080/sse",
+        "https://api.example.com/sse",
+        "http://localhost:8080/mcp",
+        "https://api.example.com/mcp",
+    ]
+    for url in urls:
+        transport = get_transport_type(url)
+        legacy = is_legacy_sse_url(url)
+        print(f"   {url}")
+        print(f"      Transport: {transport}, Legacy: {legacy}")
+    
+    # Transport negotiation
+    print("\n2. Transport Negotiation (HTTP response based):")
+    responses = [
+        (200, "application/json", "Streamable HTTP"),
+        (200, "text/event-stream", "SSE stream"),
+        (400, None, "Fallback to SSE"),
+        (404, None, "Fallback to SSE"),
+        (405, None, "Fallback to SSE"),
+    ]
+    for status, content_type, desc in responses:
+        detected = detect_transport_support(status, content_type)
+        print(f"   HTTP {status} ({desc}): {detected}")
+    
+    # Protocol versions
+    print("\n3. Supported Protocol Versions:")
+    for version in sorted(VALID_PROTOCOL_VERSIONS):
+        print(f"   - {version}")
+    
+    # Negotiator
+    print("\n4. Transport Negotiator:")
+    negotiator = TransportNegotiator("https://api.example.com/mcp")
+    print(f"   Base URL: {negotiator.base_url}")
+    print(f"   Try Streamable first: {negotiator.should_try_streamable_first()}")
+    
+    negotiator_sse = TransportNegotiator("https://api.example.com/sse")
+    print(f"   SSE URL: {negotiator_sse.base_url}")
+    print(f"   Try Streamable first: {negotiator_sse.should_try_streamable_first()}")

examples/python/mcp/mcp-custom-transport.py

@@ -0,0 +1,110 @@
+"""
+MCP Custom Transport Example
+
+Demonstrates how to create custom MCP transports by extending
+the BaseTransport class.
+
+Features:
+- BaseTransport abstract class
+- TransportRegistry for registration
+- TransportConfig for configuration
+- Pluggable transport mechanism
+
+Protocol: MCP 2025-11-25
+"""
+
+from praisonaiagents.mcp.mcp_transport import (
+    BaseTransport,
+    TransportConfig,
+    TransportRegistry,
+    get_default_registry
+)
+
+# Custom transport implementation
+class MyCustomTransport(BaseTransport):
+    """Example custom transport implementation."""
+    
+    def __init__(self, url: str, config: TransportConfig = None):
+        self.url = url
+        self.config = config or TransportConfig()
+        self._connected = False
+    
+    async def connect(self):
+        """Establish connection."""
+        print(f"Connecting to {self.url}...")
+        self._connected = True
+        print("Connected!")
+    
+    async def send(self, message: dict):
+        """Send JSON-RPC message."""
+        if not self._connected:
+            raise RuntimeError("Not connected")
+        print(f"Sending: {message}")
+    
+    async def receive(self) -> dict:
+        """Receive JSON-RPC message."""
+        if not self._connected:
+            raise RuntimeError("Not connected")
+        return {"jsonrpc": "2.0", "result": "ok", "id": 1}
+    
+    async def close(self):
+        """Close connection."""
+        print("Closing connection...")
+        self._connected = False
+    
+    @property
+    def is_connected(self) -> bool:
+        return self._connected
+
+
+if __name__ == "__main__":
+    import asyncio
+    
+    print("MCP Custom Transport Example")
+    print("=" * 50)
+    
+    # 1. Transport configuration
+    print("\n1. Transport Configuration:")
+    config = TransportConfig(
+        timeout=60,
+        debug=True,
+        retry_count=3,
+        retry_delay=1.0,
+        auth_token="my-token"
+    )
+    print(f"   Timeout: {config.timeout}s")
+    print(f"   Debug: {config.debug}")
+    print(f"   Retry count: {config.retry_count}")
+    
+    # 2. Default registry
+    print("\n2. Default Transport Registry:")
+    registry = get_default_registry()
+    print(f"   Registered transports: {registry.list_transports()}")
+    
+    # 3. Register custom transport
+    print("\n3. Registering Custom Transport:")
+    custom_registry = TransportRegistry()
+    custom_registry.register("my_custom", MyCustomTransport)
+    print(f"   Registered: {custom_registry.list_transports()}")
+    
+    # 4. Use custom transport
+    print("\n4. Using Custom Transport:")
+    
+    async def demo():
+        transport = MyCustomTransport("custom://localhost:8080", config)
+        
+        async with transport:
+            await transport.send({"jsonrpc": "2.0", "method": "test", "id": 1})
+            response = await transport.receive()
+            print(f"   Response: {response}")
+    
+    asyncio.run(demo())
+    
+    # 5. BaseTransport interface
+    print("\n5. BaseTransport Interface:")
+    print("   Required methods:")
+    print("   - async connect()")
+    print("   - async send(message: dict)")
+    print("   - async receive() -> dict")
+    print("   - async close()")
+    print("   - @property is_connected -> bool")

examples/python/mcp/mcp-resumability.py

@@ -0,0 +1,47 @@
+"""
+MCP Resumability Example
+
+Demonstrates SSE stream resumability for MCP transports.
+Resumability allows recovering from disconnections without losing messages.
+
+Features:
+- Event ID tracking per stream
+- Last-Event-ID header on reconnection
+- Retry delay handling from server
+- Message replay on reconnection
+
+Protocol: MCP 2025-11-25
+"""
+
+from praisonaiagents import Agent, MCP
+
+# Agent with resumability support (automatic)
+agent = Agent(
+    name="Resumable Assistant",
+    instructions="You are a helpful assistant with resumable connections.",
+    tools=MCP("https://api.example.com/mcp")
+)
+
+# Resumability is automatic - the MCP client:
+# 1. Tracks event IDs from SSE streams (id: field)
+# 2. Stores last event ID per stream
+# 3. Includes Last-Event-ID header on reconnection
+# 4. Respects retry delay from server (retry: field)
+
+if __name__ == "__main__":
+    print("MCP Resumability Example")
+    print("=" * 40)
+    print()
+    print("SSE Event Format:")
+    print("  id: event-123")
+    print("  retry: 3000")
+    print("  data: {\"jsonrpc\": \"2.0\", ...}")
+    print()
+    print("Resumability Flow:")
+    print("1. Client receives SSE events with id: field")
+    print("2. Client stores last event ID")
+    print("3. On disconnection, client reconnects with Last-Event-ID header")
+    print("4. Server replays messages after that event ID")
+    print("5. Client respects retry: field for reconnection delay")
+    print()
+    print("This prevents message loss during network interruptions.")

examples/python/mcp/mcp-schema-converter.py

@@ -0,0 +1,82 @@
+"""MCP Schema Converter Example
+
+Demonstrates how to convert Python functions to MCP-compatible schemas.
+
+Usage:
+    python mcp-schema-converter.py
+"""
+
+from typing import List, Dict, Any, Optional
+from praisonaiagents.mcp import function_to_mcp_schema, get_tool_metadata
+
+
+# Example functions with various type annotations
+
+def simple_search(query: str) -> str:
+    """Search for information."""
+    return f"Results for {query}"
+
+
+def advanced_search(
+    query: str,
+    max_results: int = 10,
+    include_images: bool = False,
+    filters: Optional[Dict[str, str]] = None
+) -> Dict[str, Any]:
+    """Advanced search with multiple options.
+    
+    Args:
+        query: The search query
+        max_results: Maximum results to return
+        include_images: Whether to include images
+        filters: Optional filters to apply
+    """
+    return {}
+
+
+def process_items(items: List[str], batch_size: int = 5) -> List[Dict[str, Any]]:
+    """Process a list of items.
+    
+    Args:
+        items: List of items to process
+        batch_size: Size of each batch
+    """
+    return []
+
+
+async def async_fetch(url: str, timeout: float = 30.0) -> Dict[str, Any]:
+    """Fetch data from URL asynchronously.
+    
+    Args:
+        url: The URL to fetch
+        timeout: Request timeout in seconds
+    """
+    return {}
+
+
+def main():
+    functions = [simple_search, advanced_search, process_items, async_fetch]
+    
+    print("=" * 60)
+    print("MCP Schema Converter Demo")
+    print("=" * 60)
+    
+    for func in functions:
+        print(f"\n### Function: {func.__name__}")
+        
+        # Get metadata
+        metadata = get_tool_metadata(func)
+        print(f"Name: {metadata['name']}")
+        print(f"Description: {metadata['description']}")
+        
+        # Get full schema
+        schema = function_to_mcp_schema(func)
+        
+        print("\nMCP Schema:")
+        import json
+        print(json.dumps(schema, indent=2))
+        print("-" * 40)
+
+
+if __name__ == "__main__":
+    main()

examples/python/mcp/mcp-security.py

@@ -0,0 +1,79 @@
+"""
+MCP Security Example
+
+Demonstrates security features for MCP transports.
+These features help prevent attacks like DNS rebinding.
+
+Features:
+- Origin header validation
+- DNS rebinding prevention
+- Localhost binding for local servers
+- Authentication header support
+- Secure session ID generation
+
+Protocol: MCP 2025-11-25
+"""
+
+# Example code (requires running MCP servers):
+#
+# from praisonaiagents import Agent, MCP
+#
+# # Agent with authentication
+# agent = Agent(
+#     name="Secure Assistant",
+#     tools=MCP(
+#         "https://api.example.com/mcp",
+#         headers={"Authorization": "Bearer your-token"}
+#     )
+# )
+#
+# # WebSocket with auth token
+# agent_ws = Agent(
+#     name="Secure WebSocket Assistant",
+#     tools=MCP(
+#         "wss://api.example.com/mcp",
+#         auth_token="Bearer your-secret-token"
+#     )
+# )
+
+# Using security utilities directly (no server required)
+if __name__ == "__main__":
+    from praisonaiagents.mcp.mcp_security import (
+        is_valid_origin,
+        create_auth_header,
+        generate_secure_session_id,
+        SecurityConfig
+    )
+    
+    print("MCP Security Example")
+    print("=" * 40)
+    
+    # Origin validation (DNS rebinding prevention)
+    print("\n1. Origin Validation:")
+    allowed = ["localhost", "127.0.0.1", "example.com"]
+    print(f"   Allowed origins: {allowed}")
+    print(f"   'http://localhost:8080' valid: {is_valid_origin('http://localhost:8080', allowed)}")
+    print(f"   'https://evil.com' valid: {is_valid_origin('https://evil.com', allowed)}")
+    
+    # Authentication headers
+    print("\n2. Authentication Headers:")
+    bearer = create_auth_header("my-token", auth_type="bearer")
+    print(f"   Bearer: {bearer}")
+    basic = create_auth_header("user:pass", auth_type="basic")
+    print(f"   Basic: {basic}")
+    
+    # Secure session IDs
+    print("\n3. Secure Session IDs:")
+    session_id = generate_secure_session_id()
+    print(f"   Generated: {session_id}")
+    print(f"   Length: {len(session_id)} chars")
+    
+    # Security config
+    print("\n4. Security Configuration:")
+    config = SecurityConfig(
+        allowed_origins=["localhost", "example.com"],
+        require_auth=True
+    )
+    print(f"   Validate origin: {config.validate_origin}")
+    print(f"   Require auth: {config.require_auth}")
+    print(f"   Bind address: {config.get_bind_address()}")

examples/python/mcp/mcp-session-management.py

@@ -0,0 +1,44 @@
+"""
+MCP Session Management Example
+
+Demonstrates session management features for MCP transports.
+Sessions allow stateful interactions with MCP servers.
+
+Features:
+- Automatic session ID handling
+- Session expiration detection (HTTP 404)
+- Session termination via HTTP DELETE
+- Protocol version negotiation
+
+Protocol: MCP 2025-11-25
+"""
+
+from praisonaiagents import Agent, MCP
+
+# Basic agent with automatic session management
+agent = Agent(
+    name="Session Assistant",
+    instructions="You are a helpful assistant with session support.",
+    tools=MCP("https://api.example.com/mcp")
+)
+
+# Session management is automatic - the MCP client:
+# 1. Receives Mcp-Session-Id from server during initialization
+# 2. Includes session ID in all subsequent requests
+# 3. Handles HTTP 404 (session expired) by reinitializing
+# 4. Can terminate session explicitly when done
+
+if __name__ == "__main__":
+    print("MCP Session Management Example")
+    print("=" * 40)
+    print()
+    print("Session Lifecycle:")
+    print("1. Client sends InitializeRequest")
+    print("2. Server returns InitializeResult with Mcp-Session-Id header")
+    print("3. Client includes Mcp-Session-Id in all subsequent requests")
+    print("4. If HTTP 404 received, client reinitializes session")
+    print("5. Client can send HTTP DELETE to terminate session")
+    print()
+    print("Headers used:")
+    print("- Mcp-Session-Id: Session identifier")
+    print("- Mcp-Protocol-Version: Protocol version (e.g., 2025-03-26)")

examples/python/mcp/mcp-tools-server-async.py

@@ -0,0 +1,79 @@
+"""MCP Tools Server with Async Functions
+
+Example showing how to expose async Python functions as MCP tools.
+
+Usage:
+    python mcp-tools-server-async.py
+"""
+
+import asyncio
+from praisonaiagents.mcp import ToolsMCPServer
+
+
+async def async_search(query: str, delay: float = 0.1) -> dict:
+    """Async search function.
+    
+    Args:
+        query: The search query
+        delay: Simulated delay in seconds
+    """
+    await asyncio.sleep(delay)  # Simulate async operation
+    return {
+        "query": query,
+        "results": [f"Async result for {query}"]
+    }
+
+
+async def async_fetch_url(url: str) -> dict:
+    """Fetch content from a URL asynchronously.
+    
+    Args:
+        url: The URL to fetch
+    """
+    # Mock implementation - use aiohttp in production
+    await asyncio.sleep(0.1)
+    return {
+        "url": url,
+        "status": 200,
+        "content": f"Content from {url}"
+    }
+
+
+async def async_process_data(data: str, iterations: int = 3) -> dict:
+    """Process data asynchronously with multiple iterations.
+    
+    Args:
+        data: The data to process
+        iterations: Number of processing iterations
+    """
+    results = []
+    for i in range(iterations):
+        await asyncio.sleep(0.05)
+        results.append(f"Iteration {i+1}: processed {data}")
+    
+    return {
+        "input": data,
+        "iterations": iterations,
+        "results": results
+    }
+
+
+# Sync function can also be mixed with async
+def sync_calculate(a: int, b: int) -> dict:
+    """Synchronous calculation (can be mixed with async tools)."""
+    return {"sum": a + b, "product": a * b}
+
+
+if __name__ == "__main__":
+    server = ToolsMCPServer(name="async-tools-server")
+    
+    # Register both async and sync tools
+    server.register_tools([
+        async_search,
+        async_fetch_url,
+        async_process_data,
+        sync_calculate
+    ])
+    
+    print(f"Registered tools: {server.get_tool_names()}")
+    server.run()

examples/python/mcp/mcp-tools-server-builtin.py

@@ -0,0 +1,27 @@
+"""MCP Server with Built-in PraisonAI Tools
+
+Expose PraisonAI's built-in tools as an MCP server.
+
+Usage:
+    # Set required API keys first
+    export TAVILY_API_KEY=your_key
+    
+    python mcp-tools-server-builtin.py
+"""
+
+from praisonaiagents.mcp import launch_tools_mcp_server
+
+
+if __name__ == "__main__":
+    # Expose built-in PraisonAI tools as MCP server
+    # This allows external MCP clients to use these tools
+    
+    launch_tools_mcp_server(
+        tool_names=[
+            "tavily_search",      # Web search via Tavily
+            "duckduckgo_search",  # Web search via DuckDuckGo
+            # Add more built-in tools as needed
+        ],
+        name="praisonai-builtin-tools",
+        transport="stdio"  # Use "sse" for web clients
+    )

examples/python/mcp/mcp-tools-server-simple.py

@@ -0,0 +1,25 @@
+"""Simple MCP Tools Server Example
+
+Minimal example of exposing Python functions as MCP tools.
+
+Usage:
+    python mcp-tools-server-simple.py
+"""
+
+from praisonaiagents.mcp import ToolsMCPServer
+
+
+def search(query: str) -> dict:
+    """Search for information."""
+    return {"query": query, "results": [f"Result for {query}"]}
+
+
+def add(a: int, b: int) -> dict:
+    """Add two numbers."""
+    return {"result": a + b}
+
+
+if __name__ == "__main__":
+    server = ToolsMCPServer(name="simple-tools")
+    server.register_tools([search, add])
+    server.run()

examples/python/mcp/mcp-tools-server-sse.py

@@ -0,0 +1,65 @@
+"""MCP Tools Server with SSE Transport
+
+Expose Python functions as MCP tools via Server-Sent Events (SSE).
+This is useful for web-based clients and remote access.
+
+Usage:
+    python mcp-tools-server-sse.py
+    
+    # Then connect from another agent:
+    # agent = Agent(tools=MCP("http://localhost:8080/sse"))
+"""
+
+from praisonaiagents.mcp import ToolsMCPServer
+
+
+def search_web(query: str, max_results: int = 5) -> dict:
+    """Search the web for information.
+    
+    Args:
+        query: The search query
+        max_results: Maximum results to return
+    """
+    return {
+        "query": query,
+        "results": [
+            {"title": f"Result {i+1}", "url": f"https://example.com/{i+1}"}
+            for i in range(max_results)
+        ]
+    }
+
+
+def calculate(expression: str) -> dict:
+    """Evaluate a mathematical expression.
+    
+    Args:
+        expression: Math expression (e.g., "2 + 2 * 3")
+    """
+    try:
+        allowed = set("0123456789+-*/.() ")
+        if not all(c in allowed for c in expression):
+            return {"error": "Invalid characters"}
+        return {"expression": expression, "result": eval(expression)}
+    except Exception as e:
+        return {"error": str(e)}
+
+
+def get_time() -> dict:
+    """Get the current time."""
+    from datetime import datetime
+    now = datetime.now()
+    return {
+        "time": now.strftime("%H:%M:%S"),
+        "date": now.strftime("%Y-%m-%d"),
+        "timestamp": now.isoformat()
+    }
+
+
+if __name__ == "__main__":
+    server = ToolsMCPServer(name="sse-tools-server")
+    server.register_tools([search_web, calculate, get_time])
+    
+    print("Starting SSE MCP Server...")
+    print("Connect with: MCP('http://localhost:8080/sse')")
+    
+    server.run_sse(host="0.0.0.0", port=8080)

examples/python/mcp/mcp-tools-server.py

@@ -0,0 +1,140 @@
+"""MCP Tools Server Example
+
+Expose Python functions as MCP tools for external clients like Claude Desktop, Cursor, etc.
+
+Usage:
+    # Run as stdio server (for Claude Desktop)
+    python mcp-tools-server.py
+    
+    # Run as SSE server (for web clients)
+    python mcp-tools-server.py --sse --port 8080
+"""
+
+import argparse
+from typing import List, Dict, Any
+
+
+# Define your custom tools as regular Python functions
+# Type hints and docstrings are used to generate MCP schemas
+
+def search_web(query: str, max_results: int = 5) -> Dict[str, Any]:
+    """Search the web for information.
+    
+    Args:
+        query: The search query string
+        max_results: Maximum number of results to return (default: 5)
+    
+    Returns:
+        Search results with titles and URLs
+    """
+    # Mock implementation - replace with actual search logic
+    return {
+        "query": query,
+        "results": [
+            {"title": f"Result {i+1} for '{query}'", "url": f"https://example.com/{i+1}"}
+            for i in range(max_results)
+        ]
+    }
+
+
+def calculate(expression: str) -> Dict[str, Any]:
+    """Evaluate a mathematical expression.
+    
+    Args:
+        expression: A mathematical expression to evaluate (e.g., "2 + 2 * 3")
+    
+    Returns:
+        The result of the calculation
+    """
+    try:
+        # Safe evaluation of mathematical expressions
+        allowed_chars = set("0123456789+-*/.() ")
+        if not all(c in allowed_chars for c in expression):
+            return {"error": "Invalid characters in expression"}
+        
+        result = eval(expression)  # Use safer alternative in production
+        return {"expression": expression, "result": result}
+    except Exception as e:
+        return {"error": str(e)}
+
+
+def get_weather(city: str, units: str = "celsius") -> Dict[str, Any]:
+    """Get current weather for a city.
+    
+    Args:
+        city: Name of the city
+        units: Temperature units - "celsius" or "fahrenheit" (default: celsius)
+    
+    Returns:
+        Weather information for the city
+    """
+    # Mock implementation - replace with actual weather API
+    return {
+        "city": city,
+        "temperature": 22 if units == "celsius" else 72,
+        "units": units,
+        "condition": "Sunny",
+        "humidity": 45
+    }
+
+
+def list_files(directory: str = ".", pattern: str = "*") -> List[str]:
+    """List files in a directory.
+    
+    Args:
+        directory: Directory path to list (default: current directory)
+        pattern: Glob pattern to filter files (default: *)
+    
+    Returns:
+        List of file names matching the pattern
+    """
+    import glob
+    import os
+    
+    try:
+        full_pattern = os.path.join(directory, pattern)
+        files = glob.glob(full_pattern)
+        return [os.path.basename(f) for f in files]
+    except Exception as e:
+        return [f"Error: {str(e)}"]
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Run PraisonAI Tools as MCP Server")
+    parser.add_argument("--sse", action="store_true", help="Use SSE transport instead of stdio")
+    parser.add_argument("--port", type=int, default=8080, help="Port for SSE server (default: 8080)")
+    parser.add_argument("--host", type=str, default="0.0.0.0", help="Host for SSE server")
+    parser.add_argument("--debug", action="store_true", help="Enable debug logging")
+    args = parser.parse_args()
+    
+    from praisonaiagents.mcp import ToolsMCPServer
+    
+    # Create server with custom name
+    server = ToolsMCPServer(
+        name="praisonai-custom-tools",
+        debug=args.debug
+    )
+    
+    # Register all tools
+    server.register_tools([
+        search_web,
+        calculate,
+        get_weather,
+        list_files
+    ])
+    
+    # Print registered tools
+    print(f"📦 Registered {len(server.tools)} tools:")
+    for name in server.get_tool_names():
+        print(f"   - {name}")
+    print()
+    
+    # Run the server
+    if args.sse:
+        server.run_sse(host=args.host, port=args.port)
+    else:
+        server.run_stdio()
+
+
+if __name__ == "__main__":
+    main()

examples/python/mcp/mcp-transports-overview.py

@@ -0,0 +1,91 @@
+"""
+MCP Transports Overview Example
+
+Demonstrates all MCP transport types and auto-detection.
+The MCP class automatically selects the appropriate transport
+based on the URL pattern.
+
+Transport Types:
+- stdio: Command strings (local subprocess)
+- Streamable HTTP: http:// or https:// URLs
+- WebSocket: ws:// or wss:// URLs
+- SSE (Legacy): URLs ending in /sse
+
+Protocol: MCP 2025-11-25
+"""
+
+# Example code (requires running MCP servers):
+#
+# from praisonaiagents import Agent, MCP
+#
+# # 1. stdio Transport - Local subprocess
+# agent_stdio = Agent(
+#     name="Stdio Assistant",
+#     tools=MCP("npx @modelcontextprotocol/server-memory")
+# )
+#
+# # 2. Streamable HTTP Transport
+# agent_http = Agent(
+#     name="HTTP Assistant",
+#     tools=MCP("https://api.example.com/mcp")
+# )
+#
+# # 3. WebSocket Transport
+# agent_ws = Agent(
+#     name="WebSocket Assistant",
+#     tools=MCP("wss://api.example.com/mcp")
+# )
+#
+# # 4. SSE Transport (Legacy)
+# agent_sse = Agent(
+#     name="SSE Assistant",
+#     tools=MCP("http://localhost:8080/sse")
+# )
+
+if __name__ == "__main__":
+    from praisonaiagents.mcp.mcp_transport import get_transport_type
+    
+    print("MCP Transports Overview")
+    print("=" * 50)
+    print()
+    print("Transport Auto-Detection:")
+    print("-" * 50)
+    
+    examples = [
+        ("npx @mcp/server-memory", "stdio"),
+        ("python3 server.py", "stdio"),
+        ("https://api.example.com/mcp", "http_stream"),
+        ("http://localhost:8080/mcp", "http_stream"),
+        ("ws://localhost:8080/mcp", "websocket"),
+        ("wss://api.example.com/mcp", "websocket"),
+        ("http://localhost:8080/sse", "sse"),
+        ("https://api.example.com/sse", "sse"),
+    ]
+    
+    for url, expected in examples:
+        detected = get_transport_type(url)
+        status = "✓" if detected == expected else "✗"
+        print(f"  {status} '{url}'")
+        print(f"      → {detected}")
+    
+    print()
+    print("Transport Features:")
+    print("-" * 50)
+    print("  stdio:")
+    print("    - Local subprocess communication")
+    print("    - Newline-delimited JSON-RPC")
+    print("    - Environment variable support")
+    print()
+    print("  Streamable HTTP:")
+    print("    - Session management (Mcp-Session-Id)")
+    print("    - Protocol versioning (Mcp-Protocol-Version)")
+    print("    - SSE streaming with resumability")
+    print()
+    print("  WebSocket:")
+    print("    - Bidirectional real-time")
+    print("    - Auto-reconnect with exponential backoff")
+    print("    - Authentication token support")
+    print()
+    print("  SSE (Legacy):")
+    print("    - Backward compatibility")
+    print("    - Protocol version 2024-11-05")

examples/python/mcp/streamable-http-mcp.py

@@ -0,0 +1,56 @@
+"""
+MCP Streamable HTTP Transport Example
+
+Demonstrates connecting to MCP servers via Streamable HTTP transport.
+This is the current standard transport (Protocol Revision 2025-11-25).
+
+Features:
+- Single MCP endpoint for all communication
+- Session management via Mcp-Session-Id header
+- Protocol versioning via Mcp-Protocol-Version header
+- SSE streaming with resumability
+- Session termination via HTTP DELETE
+
+Protocol: MCP 2025-11-25
+"""
+
+from praisonaiagents import Agent, MCP
+
+# Basic Streamable HTTP connection
+agent = Agent(
+    name="HTTP Stream Assistant",
+    instructions="You are a helpful assistant connected via Streamable HTTP.",
+    tools=MCP("https://api.example.com/mcp")
+)
+
+# With authentication headers
+agent_auth = Agent(
+    name="Authenticated Assistant",
+    instructions="You are a helpful assistant with authenticated connection.",
+    tools=MCP(
+        "https://api.example.com/mcp",
+        headers={"Authorization": "Bearer your-token"}
+    )
+)
+
+# With timeout and debug
+agent_debug = Agent(
+    name="Debug Assistant",
+    tools=MCP(
+        "https://api.example.com/mcp",
+        timeout=120,
+        debug=True
+    )
+)
+
+if __name__ == "__main__":
+    print("Streamable HTTP MCP Transport Example")
+    print("=" * 40)
+    print("This example requires a Streamable HTTP MCP server running.")
+    print("URL pattern: https://host:port/mcp")
+    print()
+    print("Features enabled:")
+    print("- Mcp-Protocol-Version header (default: 2025-03-26)")
+    print("- Mcp-Session-Id header (automatic)")
+    print("- SSE streaming support")
+    print("- Resumability via Last-Event-ID")

examples/python/mcp/websocket-mcp.py

@@ -0,0 +1,55 @@
+"""
+MCP WebSocket Transport Example
+
+Demonstrates connecting to MCP servers via WebSocket (ws:// or wss://).
+WebSocket transport provides bidirectional real-time communication,
+ideal for cloud deployments and long-lived connections.
+
+Features:
+- Auto-detection of ws:// and wss:// URLs
+- Authentication token support
+- Auto-reconnect with exponential backoff
+- Ping/pong keepalive
+
+Protocol: SEP-1288 (WebSocket Transport)
+"""
+
+from praisonaiagents import Agent, MCP
+
+# Basic WebSocket connection
+agent = Agent(
+    name="WebSocket Assistant",
+    instructions="You are a helpful assistant connected via WebSocket.",
+    tools=MCP("ws://localhost:8080/mcp")
+)
+
+# Secure WebSocket with authentication
+agent_secure = Agent(
+    name="Secure WebSocket Assistant",
+    instructions="You are a helpful assistant with secure WebSocket connection.",
+    tools=MCP(
+        "wss://api.example.com/mcp",
+        auth_token="Bearer your-secret-token",
+        timeout=60
+    )
+)
+
+# WebSocket with custom options
+agent_custom = Agent(
+    name="Custom WebSocket Assistant",
+    tools=MCP(
+        "wss://api.example.com/mcp",
+        auth_token="your-token",
+        timeout=120,
+        debug=True
+    )
+)
+
+if __name__ == "__main__":
+    # Note: Requires a running WebSocket MCP server
+    # Example server: https://github.com/modelcontextprotocol/servers
+    
+    print("WebSocket MCP Transport Example")
+    print("=" * 40)
+    print("This example requires a WebSocket MCP server running.")
+    print("URL patterns: ws://host:port/path or wss://host:port/path")

examples/python/workflows/simple_workflow.py

@@ -0,0 +1,38 @@
+"""
+Simple Agentic Workflow Example
+
+Demonstrates using Agent instances directly as workflow steps.
+Each agent processes the input and passes output to the next agent.
+"""
+
+from praisonaiagents import Agent, Workflow
+
+# Create agents with specific roles
+researcher = Agent(
+    name="Researcher",
+    role="Research Analyst",
+    goal="Research and provide information about topics",
+    instructions="You are a research analyst. Provide concise, factual information."
+)
+
+writer = Agent(
+    name="Writer", 
+    role="Content Writer",
+    goal="Write engaging content based on research",
+    instructions="You are a content writer. Write clear, engaging content based on the research provided."
+)
+
+# Create workflow with agents as steps
+workflow = Workflow(
+    name="Simple Agentic Pipeline",
+    steps=[researcher, writer]
+)
+
+if __name__ == "__main__":
+    # Run workflow - agents process sequentially
+    result = workflow.start(
+        "What are the key benefits of AI agents?",
+        verbose=True
+    )
+    
+    print(f"\nFinal output:\n{result['output']}")

examples/python/workflows/simple_workflow_non_agentic.py

@@ -0,0 +1,36 @@
+"""
+Simple Workflow Example
+
+The easiest way to create and run workflows in PraisonAI.
+Just pass functions or agents as steps - no complex configuration needed!
+"""
+
+from praisonaiagents import Workflow, WorkflowContext, StepResult
+
+# Define simple handler functions
+def validate(ctx: WorkflowContext) -> StepResult:
+    """Validate the input data."""
+    if not ctx.input:
+        return StepResult(output="No input provided", stop_workflow=True)
+    return StepResult(output=f"Validated: {ctx.input}")
+
+def process(ctx: WorkflowContext) -> StepResult:
+    """Process the validated data."""
+    return StepResult(output=f"Processed: {ctx.previous_result}")
+
+def format_output(ctx: WorkflowContext) -> StepResult:
+    """Format the final output."""
+    return StepResult(output=f"✅ Final: {ctx.previous_result}")
+
+# Create workflow - just list your functions!
+workflow = Workflow(
+    name="Simple Pipeline",
+    steps=[validate, process, format_output]
+)
+
+if __name__ == "__main__":
+    # Run the workflow - PraisonAI style uses .start()
+    result = workflow.start("Hello, World!", verbose=True)
+    
+    print(f"\nFinal output: {result['output']}")
+    print(f"Steps completed: {len(result['steps'])}")

examples/python/workflows/task_callbacks.py

@@ -0,0 +1,57 @@
+"""
+Task Callbacks Example
+
+Demonstrates using on_task_start and on_task_complete callbacks
+with PraisonAIAgents for workflow monitoring and logging.
+"""
+
+from praisonaiagents import Agent, Task, PraisonAIAgents
+
+# Create an agent
+researcher = Agent(
+    name="Researcher",
+    role="Research Assistant",
+    goal="Find and summarize information",
+    llm="gpt-4o-mini"
+)
+
+# Create tasks with variables
+task1 = Task(
+    name="research",
+    description="Research the topic: {{topic}}",
+    expected_output="A brief summary of the topic",
+    agent=researcher,
+    variables={"topic": "artificial intelligence"}
+)
+
+task2 = Task(
+    name="summarize",
+    description="Create a one-paragraph summary of the research.",
+    expected_output="A concise summary",
+    agent=researcher
+)
+
+# Define callbacks
+def on_start(task, task_id):
+    """Called before each task starts."""
+    print(f"🚀 Starting task: {task.name} (ID: {task_id})")
+
+def on_complete(task, output):
+    """Called after each task completes."""
+    print(f"✅ Completed task: {task.name}")
+    print(f"   Output preview: {str(output)[:100]}...")
+
+if __name__ == "__main__":
+    # Create agents with callbacks
+    agents = PraisonAIAgents(
+        agents=[researcher],
+        tasks=[task1, task2],
+        process="workflow",
+        on_task_start=on_start,
+        on_task_complete=on_complete,
+        variables={"global_var": "shared_value"}  # Global variables
+    )
+    
+    # Run the workflow
+    result = agents.start()
+    print("\n🎉 Workflow completed!")

examples/python/workflows/workflow_branching.py

@@ -0,0 +1,48 @@
+"""
+Workflow Branching Example
+
+Demonstrates conditional branching in workflows where steps can route
+to different next steps based on output content.
+"""
+
+from praisonaiagents.memory.workflows import WorkflowManager, WorkflowStep, Workflow
+
+# Create a workflow with branching
+workflow = Workflow(
+    name="Decision Workflow",
+    description="A workflow that branches based on validation result",
+    steps=[
+        WorkflowStep(
+            name="validate",
+            action="Check if the number 42 is positive. Reply with 'valid' or 'invalid'.",
+            next_steps=["success_handler", "error_handler"],
+            branch_condition={
+                "valid": ["success_handler"],
+                "invalid": ["error_handler"]
+            }
+        ),
+        WorkflowStep(
+            name="success_handler",
+            action="The validation passed! Generate a success message."
+        ),
+        WorkflowStep(
+            name="error_handler",
+            action="The validation failed. Generate an error message."
+        )
+    ]
+)
+
+if __name__ == "__main__":
+    # Create manager and register workflow
+    manager = WorkflowManager()
+    manager.workflows["Decision Workflow"] = workflow
+    
+    # Execute - will branch to success_handler since 42 is positive
+    result = manager.execute(
+        "Decision Workflow",
+        default_llm="gpt-4o-mini"
+    )
+    
+    print("Workflow completed!")
+    for step_result in result["results"]:
+        print(f"  {step_result['step']}: {step_result['status']}")

examples/python/workflows/workflow_checkpoints.py

@@ -0,0 +1,63 @@
+"""
+Workflow Checkpoints Example
+
+Demonstrates saving and resuming workflow execution using checkpoints.
+Useful for long-running workflows that may be interrupted.
+"""
+
+from praisonaiagents.memory.workflows import WorkflowManager, WorkflowStep, Workflow
+
+# Create a multi-step workflow
+workflow = Workflow(
+    name="Long Process",
+    description="A workflow with checkpoints for resumability",
+    steps=[
+        WorkflowStep(
+            name="step1",
+            action="Initialize the process and prepare data."
+        ),
+        WorkflowStep(
+            name="step2", 
+            action="Process the first batch of data."
+        ),
+        WorkflowStep(
+            name="step3",
+            action="Process the second batch of data."
+        ),
+        WorkflowStep(
+            name="step4",
+            action="Finalize and generate report."
+        )
+    ]
+)
+
+if __name__ == "__main__":
+    manager = WorkflowManager()
+    manager.workflows["Long Process"] = workflow
+    
+    # Execute with checkpoint - saves after each step
+    print("=== Starting workflow with checkpoint ===")
+    result = manager.execute(
+        "Long Process",
+        default_llm="gpt-4o-mini",
+        checkpoint="my-checkpoint"  # Saves progress after each step
+    )
+    
+    print(f"Completed {len(result['results'])} steps")
+    
+    # List available checkpoints
+    print("\n=== Available checkpoints ===")
+    checkpoints = manager.list_checkpoints()
+    for cp in checkpoints:
+        print(f"  {cp['name']}: {cp['completed_steps']} steps completed")
+    
+    # Resume from checkpoint (if workflow was interrupted)
+    # result = manager.execute(
+    #     "Long Process",
+    #     default_llm="gpt-4o-mini",
+    #     resume="my-checkpoint"  # Resumes from last saved state
+    # )
+    
+    # Clean up checkpoint
+    manager.delete_checkpoint("my-checkpoint")
+    print("\n✅ Checkpoint deleted")

examples/python/workflows/workflow_conditional.py

@@ -0,0 +1,83 @@
+"""
+Agentic Conditional Workflow Example
+
+Demonstrates using should_run to conditionally execute agent steps
+based on input or previous results.
+"""
+
+from praisonaiagents import Agent, Workflow, WorkflowStep
+
+# Condition functions
+def is_technical(ctx) -> bool:
+    """Only run technical review for technical content."""
+    keywords = ["code", "programming", "algorithm", "database", "api"]
+    return any(keyword in ctx.input.lower() for keyword in keywords)
+
+def needs_creative(ctx) -> bool:
+    """Only run creative enhancement for creative content."""
+    keywords = ["story", "poem", "creative", "write", "narrative"]
+    return any(keyword in ctx.input.lower() for keyword in keywords)
+
+# Create agents
+analyzer = Agent(
+    name="Analyzer",
+    role="Content Analyzer",
+    goal="Analyze and process content",
+    instructions="Analyze the given content and provide insights."
+)
+
+tech_reviewer = Agent(
+    name="TechReviewer",
+    role="Technical Reviewer",
+    goal="Review technical content",
+    instructions="Review the technical aspects and provide technical feedback."
+)
+
+creative_enhancer = Agent(
+    name="CreativeEnhancer",
+    role="Creative Enhancer",
+    goal="Enhance creative content",
+    instructions="Enhance the creative aspects and make it more engaging."
+)
+
+finalizer = Agent(
+    name="Finalizer",
+    role="Content Finalizer",
+    goal="Finalize the content",
+    instructions="Provide the final polished version of the content."
+)
+
+# Create workflow with conditional steps
+workflow = Workflow(
+    name="Conditional Agentic Pipeline",
+    steps=[
+        analyzer,  # Always runs
+        WorkflowStep(
+            name="tech_review",
+            agent=tech_reviewer,
+            should_run=is_technical  # Only runs for technical content
+        ),
+        WorkflowStep(
+            name="creative_enhance",
+            agent=creative_enhancer,
+            should_run=needs_creative  # Only runs for creative content
+        ),
+        finalizer  # Always runs
+    ]
+)
+
+if __name__ == "__main__":
+    # Test 1: Technical content
+    print("=== Test 1: Technical Content ===")
+    result = workflow.start("Review this Python code for best practices", verbose=True)
+    print(f"Output: {result['output'][:200]}...\n")
+    
+    # Test 2: Creative content
+    print("=== Test 2: Creative Content ===")
+    result = workflow.start("Write a short story about a robot", verbose=True)
+    print(f"Output: {result['output'][:200]}...\n")
+    
+    # Test 3: General content (neither technical nor creative)
+    print("=== Test 3: General Content ===")
+    result = workflow.start("What is the weather like today?", verbose=True)
+    print(f"Output: {result['output'][:200]}...")

examples/python/workflows/workflow_conditional_non_agentic.py

@@ -0,0 +1,68 @@
+"""
+Workflow Conditional Execution Example
+
+Demonstrates using should_run to conditionally execute steps
+based on input or previous results.
+"""
+
+from praisonaiagents import Workflow, WorkflowStep, WorkflowContext, StepResult
+
+# Condition functions - return True to run the step, False to skip
+def is_sensitive_content(ctx: WorkflowContext) -> bool:
+    """Only run compliance check for sensitive topics."""
+    sensitive_keywords = ["legal", "medical", "financial", "security"]
+    return any(keyword in ctx.input.lower() for keyword in sensitive_keywords)
+
+def needs_translation(ctx: WorkflowContext) -> bool:
+    """Only translate if content is not in English."""
+    return ctx.variables.get("language", "en") != "en"
+
+# Handler functions
+def process_content(ctx: WorkflowContext) -> StepResult:
+    return StepResult(output=f"Processed: {ctx.input}")
+
+def compliance_check(ctx: WorkflowContext) -> StepResult:
+    return StepResult(output=f"✅ Compliance verified for: {ctx.previous_result}")
+
+def translate(ctx: WorkflowContext) -> StepResult:
+    lang = ctx.variables.get("language", "unknown")
+    return StepResult(output=f"🌐 Translated to {lang}: {ctx.previous_result}")
+
+def finalize(ctx: WorkflowContext) -> StepResult:
+    return StepResult(output=f"📄 Final: {ctx.previous_result}")
+
+# Create workflow with conditional steps
+workflow = Workflow(
+    name="Conditional Pipeline",
+    steps=[
+        WorkflowStep(name="process", handler=process_content),
+        WorkflowStep(
+            name="compliance",
+            handler=compliance_check,
+            should_run=is_sensitive_content  # Only runs for sensitive content
+        ),
+        WorkflowStep(
+            name="translate",
+            handler=translate,
+            should_run=needs_translation  # Only runs if language != "en"
+        ),
+        WorkflowStep(name="finalize", handler=finalize)
+    ]
+)
+
+if __name__ == "__main__":
+    # Test 1: Normal content (skips compliance)
+    print("=== Test 1: Normal content ===")
+    result = workflow.start("Hello world", verbose=True)
+    print(f"Output: {result['output']}\n")
+    
+    # Test 2: Sensitive content (runs compliance)
+    print("=== Test 2: Legal content ===")
+    result = workflow.start("Review this legal document", verbose=True)
+    print(f"Output: {result['output']}\n")
+    
+    # Test 3: Non-English content (runs translation)
+    print("=== Test 3: Non-English content ===")
+    workflow.variables = {"language": "es"}
+    result = workflow.start("Hola mundo", verbose=True)
+    print(f"Output: {result['output']}")

examples/python/workflows/workflow_early_stop.py

@@ -0,0 +1,69 @@
+"""
+Workflow Early Stop Example
+
+Demonstrates how to stop a workflow early using custom handler
+functions that return StepResult with stop_workflow=True.
+"""
+
+from praisonaiagents.memory.workflows import (
+    WorkflowManager, WorkflowStep, Workflow,
+    WorkflowContext, StepResult
+)
+
+# Custom validator that can stop the workflow
+def validate_data(context: WorkflowContext) -> StepResult:
+    """Check if data is valid and stop workflow if not."""
+    data = context.variables.get("data", {})
+    
+    if data.get("value", 0) < 0:
+        return StepResult(
+            output="❌ Validation failed: negative value detected. Stopping workflow.",
+            stop_workflow=True  # This stops the entire workflow
+        )
+    
+    return StepResult(
+        output="✅ Validation passed: data is valid.",
+        stop_workflow=False  # Continue to next step
+    )
+
+# Create workflow with early stop capability
+workflow = Workflow(
+    name="Data Processing",
+    description="Process data with validation gate",
+    variables={
+        "data": {"value": -5}  # Invalid data - will trigger early stop
+    },
+    steps=[
+        WorkflowStep(
+            name="validate",
+            handler=validate_data  # Custom function
+        ),
+        WorkflowStep(
+            name="process",
+            action="Process the validated data."  # Won't run if validation fails
+        ),
+        WorkflowStep(
+            name="report",
+            action="Generate final report."  # Won't run if validation fails
+        )
+    ]
+)
+
+if __name__ == "__main__":
+    manager = WorkflowManager()
+    manager.workflows["Data Processing"] = workflow
+    
+    # Test with invalid data (will stop early)
+    print("=== Testing with invalid data ===")
+    result = manager.execute("Data Processing", default_llm="gpt-4o-mini")
+    
+    for step_result in result["results"]:
+        print(f"  {step_result['step']}: {step_result['output']}")
+    
+    # Test with valid data (will complete)
+    print("\n=== Testing with valid data ===")
+    workflow.variables["data"] = {"value": 42}
+    result = manager.execute("Data Processing", default_llm="gpt-4o-mini")
+    
+    for step_result in result["results"]:
+        print(f"  {step_result['step']}: {step_result['status']}")

examples/python/workflows/workflow_loop_csv.py

@@ -0,0 +1,59 @@
+"""
+Agentic Loop with CSV Workflow Example
+
+Demonstrates iterating over a CSV file with an agent processing each row.
+"""
+
+from praisonaiagents import Agent, Workflow
+from praisonaiagents.workflows import loop
+import tempfile
+import os
+
+# Create sample CSV file content
+csv_content = """topic,description
+AI Ethics,Ethical considerations in artificial intelligence
+Machine Learning,Algorithms that learn from data
+Neural Networks,Computing systems inspired by biological neural networks
+Natural Language Processing,AI understanding human language"""
+
+# Create processor agent
+processor = Agent(
+    name="TopicProcessor",
+    role="Topic Analyst",
+    goal="Analyze and explain topics",
+    instructions="Provide a brief, insightful analysis of the given topic."
+)
+
+# Create summarizer agent
+summarizer = Agent(
+    name="Summarizer",
+    role="Results Summarizer",
+    goal="Summarize all processed results",
+    instructions="Create a comprehensive summary of all the analyzed topics."
+)
+
+if __name__ == "__main__":
+    # Create temp CSV file
+    with tempfile.NamedTemporaryFile(mode='w', suffix='.csv', delete=False) as f:
+        f.write(csv_content)
+        csv_path = f.name
+    
+    try:
+        print("=== Testing Agentic Loop with CSV ===\n")
+        print(f"Processing CSV: {csv_path}\n")
+        
+        # Create workflow with loop over CSV
+        workflow = Workflow(
+            name="CSV Topic Processor",
+            steps=[
+                loop(processor, from_csv=csv_path),
+                summarizer
+            ]
+        )
+        
+        result = workflow.start("Analyze these AI topics", verbose=True)
+        
+        print(f"\nFinal Summary:\n{result['output']}")
+        
+    finally:
+        os.unlink(csv_path)

examples/python/workflows/workflow_loop_csv_non_agentic.py

@@ -0,0 +1,55 @@
+"""
+Workflow Loop with CSV Example
+
+Demonstrates iterating over a CSV file, processing each row.
+"""
+
+from praisonaiagents import Workflow, WorkflowContext, StepResult
+from praisonaiagents.workflows import loop
+import tempfile
+import os
+
+# Create sample CSV file
+csv_content = """name,email,task
+Alice,[email protected],Write documentation
+Bob,[email protected],Review pull requests
+Charlie,[email protected],Fix critical bug
+Diana,[email protected],Deploy to production"""
+
+# Processor for each CSV row
+def process_task(ctx: WorkflowContext) -> StepResult:
+    """Process a single task from CSV."""
+    item = ctx.variables.get("item", {})
+    name = item.get("name", "Unknown")
+    task = item.get("task", "No task")
+    index = ctx.variables.get("loop_index", 0)
+    
+    return StepResult(
+        output=f"[{index+1}] {name}: {task} ✅",
+        variables={"last_processed": name}
+    )
+
+if __name__ == "__main__":
+    # Create temp CSV file
+    with tempfile.NamedTemporaryFile(mode='w', suffix='.csv', delete=False) as f:
+        f.write(csv_content)
+        csv_path = f.name
+    
+    try:
+        print("=== Testing Loop with CSV ===\n")
+        print(f"CSV file: {csv_path}\n")
+        
+        # Create workflow
+        workflow = Workflow(
+            name="CSV Task Processor",
+            steps=[loop(process_task, from_csv=csv_path)]
+        )
+        
+        result = workflow.start("Process all tasks", verbose=True)
+        
+        print(f"\nAll outputs:\n{result['output']}")
+        print(f"\nProcessed {len(result['variables']['loop_outputs'])} items")
+        print(f"Last processed: {result['variables'].get('last_processed')}")
+        
+    finally:
+        os.unlink(csv_path)

examples/python/workflows/workflow_loop_list.py

@@ -0,0 +1,52 @@
+"""
+Workflow Loop with List Example
+
+Demonstrates iterating over a list of items, processing each one.
+"""
+
+from praisonaiagents import Workflow, WorkflowContext, StepResult
+from praisonaiagents.workflows import loop
+
+# Sample data
+fruits = ["apple", "banana", "cherry", "date", "elderberry"]
+
+# Processor for each item
+def process_fruit(ctx: WorkflowContext) -> StepResult:
+    """Process a single fruit from the list."""
+    fruit = ctx.variables.get("item", "unknown")
+    index = ctx.variables.get("loop_index", 0)
+    
+    # Simulate processing
+    result = f"🍎 [{index + 1}] {fruit.upper()} - processed successfully"
+    
+    return StepResult(
+        output=result,
+        variables={"last_fruit": fruit}
+    )
+
+# Aggregator to summarize results
+def summarize(ctx: WorkflowContext) -> StepResult:
+    """Summarize all processed fruits."""
+    outputs = ctx.variables.get("loop_outputs", [])
+    summary = f"📊 Processed {len(outputs)} fruits:\n"
+    summary += "\n".join(f"  {o}" for o in outputs)
+    return StepResult(output=summary)
+
+# Create workflow
+workflow = Workflow(
+    name="Fruit Processor",
+    steps=[
+        loop(process_fruit, over="fruits"),
+        summarize
+    ],
+    variables={"fruits": fruits}
+)
+
+if __name__ == "__main__":
+    print("=== Testing Loop with List ===\n")
+    
+    result = workflow.start("Process all fruits", verbose=True)
+    
+    print(f"\nFinal Output:\n{result['output']}")
+    print(f"\nLoop outputs count: {len(result['variables']['loop_outputs'])}")
+    print(f"Last processed: {result['variables'].get('last_fruit')}")

examples/python/workflows/workflow_loops.py

@@ -0,0 +1,46 @@
+"""
+Workflow Loops Example
+
+Demonstrates iterating over a list of items in a workflow step.
+Each item is processed individually with the loop variable.
+"""
+
+from praisonaiagents.memory.workflows import WorkflowManager, WorkflowStep, Workflow
+
+# Create a workflow with loop
+workflow = Workflow(
+    name="Process Items",
+    description="Process a list of items using loop",
+    variables={
+        "items": ["apple", "banana", "cherry"]  # List to iterate over
+    },
+    steps=[
+        WorkflowStep(
+            name="process_each",
+            action="Describe the fruit: {{item}}",
+            loop_over="items",  # Variable name containing the list
+            loop_var="item"     # Variable name for current item
+        ),
+        WorkflowStep(
+            name="summarize",
+            action="Summarize all the fruits that were processed."
+        )
+    ]
+)
+
+if __name__ == "__main__":
+    # Create manager and register workflow
+    manager = WorkflowManager()
+    manager.workflows["Process Items"] = workflow
+    
+    # Execute - will process each fruit in the list
+    result = manager.execute(
+        "Process Items",
+        default_llm="gpt-4o-mini"
+    )
+    
+    print("Workflow completed!")
+    for step_result in result["results"]:
+        print(f"  {step_result['step']}: {step_result['status']}")
+        if isinstance(step_result.get("output"), list):
+            print(f"    Loop processed {len(step_result['output'])} items")

examples/python/workflows/workflow_mixed_steps.py

@@ -0,0 +1,68 @@
+"""
+Mixed Steps Workflow Example
+
+Demonstrates combining different step types in a single workflow:
+- Agent instances
+- Handler functions
+- WorkflowStep with action strings
+"""
+
+from praisonaiagents import Workflow, WorkflowStep, WorkflowContext, StepResult, Agent
+
+# Create an agent
+analyzer = Agent(
+    name="Analyzer",
+    role="Data Analyst",
+    goal="Analyze and summarize data",
+    llm="gpt-4o-mini"
+)
+
+# Custom handler functions
+def validate_input(ctx: WorkflowContext) -> StepResult:
+    """Validate the input before processing."""
+    if len(ctx.input) < 5:
+        return StepResult(
+            output="Input too short. Minimum 5 characters required.",
+            stop_workflow=True
+        )
+    return StepResult(output=f"✅ Valid input: {ctx.input}")
+
+def format_output(ctx: WorkflowContext) -> StepResult:
+    """Format the final output."""
+    return StepResult(
+        output=f"📊 Report:\n{ctx.previous_result}",
+        variables={"report_generated": True}
+    )
+
+# Create workflow with mixed step types
+workflow = Workflow(
+    name="Mixed Pipeline",
+    steps=[
+        # Step 1: Function - validates input
+        validate_input,
+        
+        # Step 2: Agent - analyzes content
+        analyzer,
+        
+        # Step 3: WorkflowStep with action string
+        WorkflowStep(
+            name="enhance",
+            action="Add a conclusion to this analysis: {{previous_output}}"
+        ),
+        
+        # Step 4: Function - formats output
+        format_output
+    ],
+    default_llm="gpt-4o-mini"  # Used for action-based steps
+)
+
+if __name__ == "__main__":
+    # Test with valid input
+    print("=== Running mixed workflow ===")
+    result = workflow.start(
+        "Analyze the trends in AI adoption for 2024",
+        verbose=True
+    )
+    
+    print(f"\nFinal output:\n{result['output']}")
+    print(f"\nVariables: {result['variables']}")

examples/python/workflows/workflow_parallel.py

@@ -0,0 +1,56 @@
+"""
+Agentic Parallel Workflow Example
+
+Demonstrates running multiple agents concurrently and
+combining their results with an aggregator agent.
+"""
+
+from praisonaiagents import Agent, Workflow
+from praisonaiagents.workflows import parallel
+
+# Create parallel research agents
+market_researcher = Agent(
+    name="MarketResearcher",
+    role="Market Research Analyst",
+    goal="Research market trends and opportunities",
+    instructions="Analyze market trends. Provide concise market insights."
+)
+
+competitor_researcher = Agent(
+    name="CompetitorResearcher", 
+    role="Competitive Intelligence Analyst",
+    goal="Research competitor strategies",
+    instructions="Analyze competitors. Provide key competitive insights."
+)
+
+customer_researcher = Agent(
+    name="CustomerResearcher",
+    role="Customer Research Analyst", 
+    goal="Research customer needs and behaviors",
+    instructions="Analyze customer segments. Provide customer insights."
+)
+
+# Create aggregator agent
+aggregator = Agent(
+    name="Aggregator",
+    role="Research Synthesizer",
+    goal="Synthesize research findings",
+    instructions="Combine all research findings into a comprehensive summary."
+)
+
+# Create workflow with parallel execution
+workflow = Workflow(
+    name="Parallel Research Pipeline",
+    steps=[
+        parallel([market_researcher, competitor_researcher, customer_researcher]),
+        aggregator
+    ]
+)
+
+if __name__ == "__main__":
+    print("=== Testing Agentic Parallel Workflow ===\n")
+    
+    # Run workflow - all researchers work in parallel, then aggregator summarizes
+    result = workflow.start("Research the AI industry", verbose=True)
+    
+    print(f"\nFinal Summary:\n{result['output']}")

examples/python/workflows/workflow_parallel_non_agentic.py

@@ -0,0 +1,53 @@
+"""
+Workflow Parallel Execution Example
+
+Demonstrates running multiple steps concurrently and
+combining their results.
+"""
+
+from praisonaiagents import Workflow, WorkflowContext, StepResult
+from praisonaiagents.workflows import parallel
+import time
+
+# Parallel workers - each does independent work
+def research_market(ctx: WorkflowContext) -> StepResult:
+    """Research market trends."""
+    time.sleep(0.1)  # Simulate work
+    return StepResult(output="📊 Market: Growth expected 15% YoY")
+
+def research_competitors(ctx: WorkflowContext) -> StepResult:
+    """Research competitor analysis."""
+    time.sleep(0.1)  # Simulate work
+    return StepResult(output="🏢 Competitors: 3 major players identified")
+
+def research_customers(ctx: WorkflowContext) -> StepResult:
+    """Research customer feedback."""
+    time.sleep(0.1)  # Simulate work
+    return StepResult(output="👥 Customers: 85% satisfaction rate")
+
+# Aggregator - combines parallel results
+def summarize_research(ctx: WorkflowContext) -> StepResult:
+    """Summarize all research findings."""
+    outputs = ctx.variables.get("parallel_outputs", [])
+    summary = "📋 RESEARCH SUMMARY:\n" + "\n".join(f"  • {o}" for o in outputs)
+    return StepResult(output=summary)
+
+# Create workflow with parallel execution
+workflow = Workflow(
+    name="Parallel Research",
+    steps=[
+        parallel([research_market, research_competitors, research_customers]),
+        summarize_research
+    ]
+)
+
+if __name__ == "__main__":
+    print("=== Testing Parallel Workflow ===\n")
+    
+    start_time = time.time()
+    result = workflow.start("Analyze the business landscape", verbose=True)
+    elapsed = time.time() - start_time
+    
+    print(f"\nFinal Output:\n{result['output']}")
+    print(f"\nCompleted in {elapsed:.2f}s (parallel execution)")
+    print(f"Individual outputs: {len(result['variables']['parallel_outputs'])}")

examples/python/workflows/workflow_repeat.py

@@ -0,0 +1,53 @@
+"""
+Agentic Evaluator-Optimizer (Repeat) Workflow Example
+
+Demonstrates the evaluator-optimizer pattern where an agent
+generates content and another evaluates it, repeating until approved.
+"""
+
+from praisonaiagents import Agent, Workflow
+from praisonaiagents.workflows import repeat
+
+# Create generator agent
+generator = Agent(
+    name="Generator",
+    role="Content Generator",
+    goal="Generate high-quality content",
+    instructions="Generate content based on the topic. Improve based on feedback if provided."
+)
+
+# Create evaluator agent  
+evaluator = Agent(
+    name="Evaluator",
+    role="Content Evaluator",
+    goal="Evaluate and provide feedback on content",
+    instructions="Evaluate the content quality. If good, respond with 'APPROVED'. Otherwise provide specific feedback for improvement."
+)
+
+# Evaluation function to check if content is approved
+def is_approved(ctx) -> bool:
+    return "approved" in ctx.previous_result.lower()
+
+# Create workflow with evaluator-optimizer pattern
+workflow = Workflow(
+    name="Evaluator-Optimizer Pipeline",
+    steps=[
+        generator,  # First, generate content
+        repeat(
+            evaluator,  # Evaluate and provide feedback
+            until=is_approved,
+            max_iterations=3
+        )
+    ]
+)
+
+if __name__ == "__main__":
+    print("=== Testing Agentic Evaluator-Optimizer Workflow ===\n")
+    
+    # Run optimization workflow
+    result = workflow.start(
+        "Write a compelling product description for an AI assistant",
+        verbose=True
+    )
+    
+    print(f"\nFinal Result:\n{result['output']}")

examples/python/workflows/workflow_repeat_non_agentic.py

@@ -0,0 +1,69 @@
+"""
+Workflow Repeat (Evaluator-Optimizer) Example
+
+Demonstrates repeating a step until a condition is met.
+This is useful for iterative improvement patterns.
+"""
+
+from praisonaiagents import Workflow, WorkflowContext, StepResult
+from praisonaiagents.workflows import repeat
+
+# Simulated content generator that improves each iteration
+class ContentGenerator:
+    def __init__(self):
+        self.iteration = 0
+        self.content = []
+    
+    def generate(self, ctx: WorkflowContext) -> StepResult:
+        """Generate content, adding more each iteration."""
+        self.iteration += 1
+        
+        # Add new content each iteration
+        new_items = [
+            f"Point {self.iteration}.1: Key insight about the topic",
+            f"Point {self.iteration}.2: Supporting evidence"
+        ]
+        self.content.extend(new_items)
+        
+        output = f"Iteration {self.iteration}: Generated {len(self.content)} points total"
+        
+        return StepResult(
+            output=output,
+            variables={
+                "total_points": len(self.content),
+                "content": self.content.copy()
+            }
+        )
+
+# Condition: stop when we have enough content
+def has_enough_content(ctx: WorkflowContext) -> bool:
+    """Check if we have generated enough content."""
+    total = ctx.variables.get("total_points", 0)
+    return total >= 6  # Stop when we have 6+ points
+
+# Create generator instance
+generator = ContentGenerator()
+
+# Create workflow with repeat pattern
+workflow = Workflow(
+    name="Content Generator",
+    steps=[
+        repeat(
+            generator.generate,
+            until=has_enough_content,
+            max_iterations=10
+        )
+    ]
+)
+
+if __name__ == "__main__":
+    print("=== Testing Repeat (Evaluator-Optimizer) Pattern ===\n")
+    
+    result = workflow.start("Generate comprehensive content", verbose=True)
+    
+    print(f"\nFinal output: {result['output']}")
+    print(f"Total iterations: {result['variables']['repeat_iterations']}")
+    print(f"Total points generated: {result['variables']['total_points']}")
+    print("\nGenerated content:")
+    for point in result['variables']['content']:
+        print(f"  • {point}")

examples/python/workflows/workflow_routing.py

@@ -0,0 +1,65 @@
+"""
+Agentic Routing Workflow Example
+
+Demonstrates decision-based routing where an agent classifier
+routes to specialized handler agents based on the request type.
+"""
+
+from praisonaiagents import Agent, Workflow
+from praisonaiagents.workflows import route
+
+# Create classifier agent
+classifier = Agent(
+    name="Classifier",
+    role="Request Classifier",
+    goal="Classify incoming requests",
+    instructions="Classify the request. Respond with ONLY 'technical', 'creative', or 'general'."
+)
+
+# Create specialized handler agents
+tech_agent = Agent(
+    name="TechExpert",
+    role="Technical Expert",
+    goal="Handle technical questions",
+    instructions="You are a technical expert. Provide detailed technical answers."
+)
+
+creative_agent = Agent(
+    name="CreativeWriter",
+    role="Creative Writer", 
+    goal="Handle creative requests",
+    instructions="You are a creative writer. Write engaging, creative content."
+)
+
+general_agent = Agent(
+    name="GeneralAssistant",
+    role="General Assistant",
+    goal="Handle general requests",
+    instructions="You are a helpful assistant. Provide clear, helpful responses."
+)
+
+# Create workflow with routing
+workflow = Workflow(
+    name="Agentic Router",
+    steps=[
+        classifier,
+        route({
+            "technical": [tech_agent],
+            "creative": [creative_agent],
+            "default": [general_agent]
+        })
+    ]
+)
+
+if __name__ == "__main__":
+    print("=== Testing Agentic Routing Workflow ===\n")
+    
+    # Test 1: Technical question
+    print("--- Technical Request ---")
+    result = workflow.start("How does machine learning work?", verbose=True)
+    print(f"Result: {result['output'][:200]}...\n")
+    
+    # Test 2: Creative request
+    print("--- Creative Request ---")
+    result = workflow.start("Write a poem about the ocean", verbose=True)
+    print(f"Result: {result['output'][:200]}...")

examples/python/workflows/workflow_routing_non_agentic.py

@@ -0,0 +1,61 @@
+"""
+Workflow Routing Example
+
+Demonstrates decision-based routing where the workflow
+takes different paths based on the output of a decision step.
+"""
+
+from praisonaiagents import Workflow, WorkflowContext, StepResult
+from praisonaiagents.workflows import route
+
+# Decision maker - determines which route to take
+def classify_request(ctx: WorkflowContext) -> StepResult:
+    """Classify the input and decide the route."""
+    input_lower = ctx.input.lower()
+    
+    if "urgent" in input_lower or "emergency" in input_lower:
+        return StepResult(output="priority: high")
+    elif "question" in input_lower or "help" in input_lower:
+        return StepResult(output="priority: support")
+    else:
+        return StepResult(output="priority: normal")
+
+# Route handlers
+def handle_high_priority(ctx: WorkflowContext) -> StepResult:
+    return StepResult(output="🚨 HIGH PRIORITY: Escalating to senior team immediately!")
+
+def handle_support(ctx: WorkflowContext) -> StepResult:
+    return StepResult(output="💬 SUPPORT: Routing to help desk for assistance.")
+
+def handle_normal(ctx: WorkflowContext) -> StepResult:
+    return StepResult(output="📋 NORMAL: Added to standard processing queue.")
+
+# Create workflow with routing
+workflow = Workflow(
+    name="Request Router",
+    steps=[
+        classify_request,
+        route({
+            "high": [handle_high_priority],
+            "support": [handle_support],
+            "normal": [handle_normal],
+            "default": [handle_normal]
+        })
+    ]
+)
+
+if __name__ == "__main__":
+    # Test different inputs
+    print("=== Testing Routing Workflow ===\n")
+    
+    # Test 1: Urgent request
+    result = workflow.start("This is an URGENT matter!", verbose=True)
+    print(f"Result: {result['output']}\n")
+    
+    # Test 2: Support request
+    result = workflow.start("I have a question about my account", verbose=True)
+    print(f"Result: {result['output']}\n")
+    
+    # Test 3: Normal request
+    result = workflow.start("Please process my order", verbose=True)
+    print(f"Result: {result['output']}")

examples/python/workflows/workflow_with_agents.py

@@ -0,0 +1,45 @@
+"""
+Workflow with Agents Example
+
+Demonstrates using Agent instances directly as workflow steps.
+Agents are automatically wrapped and executed in sequence.
+"""
+
+from praisonaiagents import Workflow, Agent
+
+# Create agents
+researcher = Agent(
+    name="Researcher",
+    role="Research Specialist",
+    goal="Find accurate information",
+    llm="gpt-4o-mini"
+)
+
+writer = Agent(
+    name="Writer", 
+    role="Content Writer",
+    goal="Write engaging content",
+    llm="gpt-4o-mini"
+)
+
+editor = Agent(
+    name="Editor",
+    role="Content Editor", 
+    goal="Polish and improve content",
+    llm="gpt-4o-mini"
+)
+
+# Create workflow with agents as steps
+workflow = Workflow(
+    name="Content Pipeline",
+    steps=[researcher, writer, editor]
+)
+
+if __name__ == "__main__":
+    # Run the workflow
+    result = workflow.start(
+        "Write a short paragraph about artificial intelligence",
+        verbose=True
+    )
+    
+    print(f"\nFinal output:\n{result['output']}")

examples/yaml/agents_workflow.yaml

@@ -0,0 +1,117 @@
+# Extended agents.yaml with Workflow Patterns
+# This file demonstrates the extended agents.yaml format that supports
+# advanced workflow patterns while maintaining backward compatibility.
+#
+# Usage: praisonai agents_workflow.yaml
+#
+# Key differences from standard agents.yaml:
+# - process: workflow (triggers workflow mode)
+# - workflow: section for planning, reasoning, memory
+# - steps: section for workflow patterns (route, parallel, loop, repeat)
+
+framework: praisonai
+process: workflow
+topic: "Research AI trends and create a comprehensive report"
+
+# Workflow configuration (new)
+workflow:
+  planning: true
+  reasoning: true
+  verbose: true
+  memory_config:
+    provider: chroma
+    persist: true
+
+# Variables for template substitution
+variables:
+  topic: AI trends
+  topics:
+    - Machine Learning
+    - Neural Networks
+    - Natural Language Processing
+
+# Agent definitions using familiar 'roles' format
+roles:
+  classifier:
+    role: Request Classifier
+    backstory: |
+      You are an expert at classifying requests into categories.
+      Respond with ONLY one word: 'technical', 'creative', or 'research'.
+    goal: Classify incoming requests accurately
+    
+  researcher:
+    role: Research Analyst
+    backstory: |
+      You are a skilled research analyst with expertise in technology trends.
+      You provide concise, factual research findings with citations.
+    goal: Research topics thoroughly and provide insights
+    tools:
+      - tavily_search
+    llm:
+      model: gpt-4o-mini
+      
+  tech_expert:
+    role: Technical Expert
+    backstory: |
+      You are a technical expert with deep knowledge in AI and ML.
+      You provide detailed technical explanations.
+    goal: Handle technical questions with expertise
+    
+  creative_writer:
+    role: Creative Writer
+    backstory: |
+      You are a creative writer who can make complex topics engaging.
+      You write in a clear, accessible style.
+    goal: Create engaging content from research
+    
+  market_analyst:
+    role: Market Analyst
+    backstory: |
+      You analyze market trends and competitive landscapes.
+      You provide actionable business insights.
+    goal: Analyze market trends and opportunities
+    
+  aggregator:
+    role: Synthesizer
+    backstory: |
+      You combine findings from multiple sources into coherent reports.
+      You identify patterns and key insights.
+    goal: Synthesize all findings into comprehensive reports
+
+# Workflow steps with patterns (new)
+steps:
+  # Step 1: Classify the request
+  - agent: classifier
+    action: "Classify this request: {{topic}}"
+    
+  # Step 2: Route to appropriate handler based on classification
+  - name: routing
+    route:
+      technical: [tech_expert]
+      creative: [creative_writer]
+      default: [researcher]
+      
+  # Step 3: Parallel research on multiple aspects
+  - name: parallel_research
+    parallel:
+      - agent: researcher
+        action: "Research technology trends for {{topic}}"
+      - agent: market_analyst
+        action: "Analyze market opportunities for {{topic}}"
+        
+  # Step 4: Loop over specific topics
+  - agent: researcher
+    action: "Deep dive into {{item}}"
+    loop:
+      over: topics
+      
+  # Step 5: Aggregate all findings
+  - agent: aggregator
+    action: "Synthesize all research findings into a comprehensive report"
+    repeat:
+      until: "comprehensive"
+      max_iterations: 2
+
+# Optional callbacks
+callbacks:
+  on_step_complete: log_step

examples/yaml/workflows/complete_workflow.yaml

@@ -0,0 +1,116 @@
+# Complete Workflow Example
+# Demonstrates all workflow patterns in a single comprehensive workflow
+
+name: Complete Research Workflow
+description: A comprehensive workflow demonstrating all patterns - route, parallel, loop, repeat
+framework: praisonai
+
+workflow:
+  planning: true
+  reasoning: true
+  verbose: true
+  memory_config:
+    provider: chroma
+    persist: true
+
+variables:
+  topic: AI trends
+  topics:
+    - Machine Learning
+    - Neural Networks
+    - Natural Language Processing
+
+agents:
+  classifier:
+    name: Classifier
+    role: Request Classifier
+    goal: Classify incoming requests
+    instructions: |
+      Classify the request. Respond with ONLY one word:
+      - 'technical' for technical questions
+      - 'creative' for creative requests
+      - 'research' for research tasks
+
+  researcher:
+    name: Researcher
+    role: Research Analyst
+    goal: Research topics thoroughly
+    instructions: "You are a research analyst. Provide concise, factual research findings."
+
+  tech_expert:
+    name: TechExpert
+    role: Technical Expert
+    goal: Handle technical questions
+    instructions: "You are a technical expert. Provide detailed technical answers."
+
+  creative_writer:
+    name: CreativeWriter
+    role: Creative Writer
+    goal: Handle creative requests
+    instructions: "You are a creative writer. Write engaging, creative content."
+
+  market_analyst:
+    name: MarketAnalyst
+    role: Market Analyst
+    goal: Analyze market trends
+    instructions: "You are a market analyst. Provide market insights."
+
+  competitor_analyst:
+    name: CompetitorAnalyst
+    role: Competitor Analyst
+    goal: Analyze competitors
+    instructions: "You are a competitor analyst. Provide competitive insights."
+
+  aggregator:
+    name: Aggregator
+    role: Synthesizer
+    goal: Combine and synthesize findings
+    instructions: "You are a synthesizer. Combine all findings into a comprehensive report."
+
+  evaluator:
+    name: Evaluator
+    role: Quality Evaluator
+    goal: Evaluate content quality
+    instructions: |
+      Evaluate the content. If it's comprehensive and well-structured, respond with "APPROVED".
+      Otherwise, provide specific feedback for improvement.
+
+steps:
+  # Step 1: Classify the request
+  - agent: classifier
+    action: "Classify: {{input}}"
+    
+  # Step 2: Route to appropriate handler
+  - name: routing
+    route:
+      technical: [tech_expert]
+      creative: [creative_writer]
+      default: [researcher]
+      
+  # Step 3: Parallel research
+  - name: parallel_research
+    parallel:
+      - agent: market_analyst
+        action: "Analyze market for {{topic}}"
+      - agent: competitor_analyst
+        action: "Analyze competitors for {{topic}}"
+        
+  # Step 4: Loop over topics
+  - agent: researcher
+    action: "Research {{item}}"
+    loop:
+      over: topics
+      
+  # Step 5: Aggregate findings
+  - agent: aggregator
+    action: "Synthesize all findings into a comprehensive report"
+    
+  # Step 6: Evaluator-optimizer
+  - agent: evaluator
+    action: "Evaluate the report quality"
+    repeat:
+      until: "approved"
+      max_iterations: 2
+
+callbacks:
+  on_step_complete: log_step

examples/yaml/workflows/evaluator_optimizer_workflow.yaml

@@ -0,0 +1,43 @@
+# Evaluator-Optimizer Workflow
+# Demonstrates the repeat() pattern for iterative improvement
+
+name: Evaluator Optimizer Workflow
+description: A generator creates content, an evaluator reviews it repeatedly until quality standards are met
+
+workflow:
+  verbose: true
+
+agents:
+  generator:
+    name: Generator
+    role: Content Generator
+    goal: Generate high-quality content
+    instructions: |
+      You are a content generator. Generate content based on the request.
+      If feedback is provided, improve your content based on that feedback.
+      Always strive for clarity, accuracy, and engagement.
+
+  evaluator:
+    name: Evaluator
+    role: Content Evaluator
+    goal: Evaluate content quality and provide feedback
+    instructions: |
+      You are a content evaluator. Review the content provided.
+      
+      If the content is high quality (clear, accurate, engaging), respond with:
+      "APPROVED: [brief reason]"
+      
+      If the content needs improvement, respond with:
+      "NEEDS IMPROVEMENT: [specific feedback for improvement]"
+      
+      Be constructive and specific in your feedback.
+
+steps:
+  - agent: generator
+    action: "Generate content for: {{input}}"
+    
+  - agent: evaluator
+    action: "Evaluate this content and provide feedback: {{previous_output}}"
+    repeat:
+      until: "approved"
+      max_iterations: 3

examples/yaml/workflows/example_usage.py

@@ -0,0 +1,152 @@
+"""
+Example usage of YAML Workflow Parser.
+
+This script demonstrates how to load and execute YAML workflow files.
+"""
+
+import os
+from pathlib import Path
+
+# Ensure OPENAI_API_KEY is set
+if not os.getenv("OPENAI_API_KEY"):
+    print("Please set OPENAI_API_KEY environment variable")
+    print("export OPENAI_API_KEY='your-api-key'")
+    exit(1)
+
+from praisonaiagents.workflows import YAMLWorkflowParser, WorkflowManager
+
+
+def example_1_parse_and_run():
+    """Example 1: Parse YAML and run workflow directly."""
+    print("=" * 60)
+    print("Example 1: Parse YAML and run workflow")
+    print("=" * 60)
+    
+    yaml_content = """
+name: Quick Research
+description: A quick research workflow
+
+agents:
+  researcher:
+    name: Researcher
+    role: Research Analyst
+    goal: Research topics
+    instructions: "Provide brief, factual information."
+
+steps:
+  - agent: researcher
+    action: "Research: {{input}}"
+"""
+    
+    parser = YAMLWorkflowParser()
+    workflow = parser.parse_string(yaml_content)
+    
+    print(f"Workflow Name: {workflow.name}")
+    print(f"Steps: {len(workflow.steps)}")
+    
+    # Run the workflow
+    result = workflow.start("What is Python?")
+    print(f"Result: {result['output'][:200]}...")
+    print()
+
+
+def example_2_load_yaml_file():
+    """Example 2: Load workflow from YAML file using WorkflowManager."""
+    print("=" * 60)
+    print("Example 2: Load YAML file with WorkflowManager")
+    print("=" * 60)
+    
+    # Get the directory of this script
+    script_dir = Path(__file__).parent
+    yaml_file = script_dir / "simple_workflow.yaml"
+    
+    if not yaml_file.exists():
+        print(f"YAML file not found: {yaml_file}")
+        return
+    
+    manager = WorkflowManager()
+    workflow = manager.load_yaml(yaml_file)
+    
+    print(f"Loaded workflow: {workflow.name}")
+    print(f"Steps: {len(workflow.steps)}")
+    
+    # Run the workflow
+    result = workflow.start("Benefits of renewable energy")
+    print(f"Result: {result['output'][:200]}...")
+    print()
+
+
+def example_3_execute_yaml_directly():
+    """Example 3: Execute YAML file directly with WorkflowManager."""
+    print("=" * 60)
+    print("Example 3: Execute YAML directly")
+    print("=" * 60)
+    
+    script_dir = Path(__file__).parent
+    yaml_file = script_dir / "simple_workflow.yaml"
+    
+    if not yaml_file.exists():
+        print(f"YAML file not found: {yaml_file}")
+        return
+    
+    manager = WorkflowManager()
+    result = manager.execute_yaml(
+        yaml_file,
+        input_data="The future of AI",
+        verbose=True
+    )
+    
+    print(f"Status: {result['status']}")
+    print(f"Result: {result['output'][:200]}...")
+    print()
+
+
+def example_4_with_variables():
+    """Example 4: YAML workflow with variables."""
+    print("=" * 60)
+    print("Example 4: Workflow with variables")
+    print("=" * 60)
+    
+    yaml_content = """
+name: Variable Workflow
+variables:
+  topic: Machine Learning
+  style: concise
+
+agents:
+  writer:
+    name: Writer
+    role: Content Writer
+    goal: Write content
+    instructions: "Write {{style}} content about the topic."
+
+steps:
+  - agent: writer
+    action: "Write about {{topic}}"
+"""
+    
+    parser = YAMLWorkflowParser()
+    workflow = parser.parse_string(yaml_content)
+    
+    print(f"Variables: {workflow.variables}")
+    
+    # Run with default variables
+    result = workflow.start("")
+    print(f"Result: {result['output'][:200]}...")
+    print()
+
+
+if __name__ == "__main__":
+    print("YAML Workflow Examples")
+    print("=" * 60)
+    print()
+    
+    # Run examples
+    example_1_parse_and_run()
+    example_2_load_yaml_file()
+    example_3_execute_yaml_directly()
+    example_4_with_variables()
+    
+    print("=" * 60)
+    print("All examples completed!")
+    print("=" * 60)

examples/yaml/workflows/loop_workflow.yaml

@@ -0,0 +1,42 @@
+# Loop Workflow
+# Demonstrates the loop() pattern for iterating over items
+
+name: Repetitive Processing Workflow
+description: An agent processes multiple items in a loop, then a summarizer combines results
+
+workflow:
+  verbose: true
+
+variables:
+  topics:
+    - Artificial Intelligence
+    - Machine Learning
+    - Neural Networks
+    - Natural Language Processing
+
+agents:
+  processor:
+    name: Processor
+    role: Topic Processor
+    goal: Process and analyze each topic thoroughly
+    instructions: |
+      You are a topic processor. For each topic given, provide:
+      1. A brief definition
+      2. Key concepts
+      3. Real-world applications
+      Keep your response concise but informative.
+
+  summarizer:
+    name: Summarizer
+    role: Results Summarizer
+    goal: Summarize all processed results into a final report
+    instructions: "You are a summarizer. Combine all the processed topics into a cohesive final report."
+
+steps:
+  - agent: processor
+    action: "Analyze this topic: {{item}}"
+    loop:
+      over: topics
+      
+  - agent: summarizer
+    action: "Create a comprehensive summary of all the analyzed topics"

examples/yaml/workflows/parallel_workflow.yaml

@@ -0,0 +1,46 @@
+# Parallel Workflow
+# Demonstrates the parallel() pattern for concurrent execution
+
+name: Parallel Research Workflow
+description: Multiple agents research different aspects concurrently, then an aggregator combines results
+
+workflow:
+  verbose: true
+
+agents:
+  market_researcher:
+    name: MarketResearcher
+    role: Market Research Analyst
+    goal: Research market trends and opportunities
+    instructions: "You are a market research analyst. Analyze market trends and provide concise insights."
+
+  competitor_researcher:
+    name: CompetitorResearcher
+    role: Competitive Intelligence Analyst
+    goal: Research competitor strategies and positioning
+    instructions: "You are a competitive intelligence analyst. Analyze competitors and provide key insights."
+
+  customer_researcher:
+    name: CustomerResearcher
+    role: Customer Research Analyst
+    goal: Research customer needs and behaviors
+    instructions: "You are a customer research analyst. Analyze customer segments and provide insights."
+
+  aggregator:
+    name: Aggregator
+    role: Research Synthesizer
+    goal: Synthesize all research findings into a comprehensive summary
+    instructions: "You are a research synthesizer. Combine all research findings into a clear, comprehensive summary."
+
+steps:
+  - name: parallel_research
+    parallel:
+      - agent: market_researcher
+        action: "Research market trends for: {{input}}"
+      - agent: competitor_researcher
+        action: "Research competitors for: {{input}}"
+      - agent: customer_researcher
+        action: "Research customer needs for: {{input}}"
+        
+  - agent: aggregator
+    action: "Synthesize all the research findings into a comprehensive report"

examples/yaml/workflows/routing_workflow.yaml

@@ -0,0 +1,49 @@
+# Routing Workflow
+# Demonstrates the route() pattern for decision-based branching
+
+name: Agentic Routing Workflow
+description: A classifier agent routes requests to specialized handler agents
+
+workflow:
+  verbose: true
+
+agents:
+  classifier:
+    name: Classifier
+    role: Request Classifier
+    goal: Classify incoming requests by type
+    instructions: |
+      You are a request classifier. Analyze the request and respond with ONLY one word:
+      - 'technical' for technical/coding questions
+      - 'creative' for creative writing requests
+      - 'general' for everything else
+      
+      Respond with ONLY the classification word, nothing else.
+
+  tech_expert:
+    name: TechExpert
+    role: Technical Expert
+    goal: Answer technical questions with expertise
+    instructions: "You are a technical expert. Provide detailed, accurate technical answers."
+
+  creative_writer:
+    name: CreativeWriter
+    role: Creative Writer
+    goal: Create engaging creative content
+    instructions: "You are a creative writer. Write engaging, imaginative content."
+
+  general_assistant:
+    name: GeneralAssistant
+    role: General Assistant
+    goal: Help with general requests
+    instructions: "You are a helpful assistant. Provide clear, helpful responses."
+
+steps:
+  - agent: classifier
+    action: "Classify this request: {{input}}"
+    
+  - name: routing
+    route:
+      technical: [tech_expert]
+      creative: [creative_writer]
+      default: [general_assistant]

examples/yaml/workflows/simple_workflow.yaml

@@ -0,0 +1,28 @@
+# Simple Sequential Workflow
+# Demonstrates basic agent-to-agent workflow
+
+name: Simple Research Workflow
+description: A simple workflow where a researcher gathers info and a writer summarizes it
+
+workflow:
+  verbose: true
+
+agents:
+  researcher:
+    name: Researcher
+    role: Research Analyst
+    goal: Research and gather information about topics
+    instructions: "You are a research analyst. Provide concise, factual information about the given topic."
+
+  writer:
+    name: Writer
+    role: Content Writer
+    goal: Write clear and engaging summaries
+    instructions: "You are a content writer. Take the research provided and write a clear, engaging summary."
+
+steps:
+  - agent: researcher
+    action: "Research the following topic: {{input}}"
+    
+  - agent: writer
+    action: "Write a summary based on this research: {{previous_output}}"

src/praisonai-agents/.gitignore

@@ -86,3 +86,4 @@ build
 graph.py
 chroma_db/
 .qodo
+*_COMPARISON_ANALYSIS.md

src/praisonai-agents/README.md

@@ -84,6 +84,64 @@ agents = PraisonAIAgents(agents=[researcher, writer], tasks=[research_task, writ
 result = agents.start()
 ```
 
+### Planning Mode 🆕
+
+#### Single Agent Planning (Simplest)
+
+Enable planning for any single agent:
+
+```python
+from praisonaiagents import Agent
+
+def search_web(query: str) -> str:
+    return f"Search results for: {query}"
+
+agent = Agent(
+    name="AI Assistant",
+    instructions="Research and write about topics",
+    planning=True,              # Enable planning mode
+    planning_tools=[search_web], # Tools for planning research
+    planning_reasoning=True      # Chain-of-thought reasoning
+)
+
+result = agent.start("Research AI trends in 2025 and write a summary")
+```
+
+#### Multi-Agent Planning
+
+Plan before execution with multiple agents:
+
+```python
+from praisonaiagents import Agent, Task, PraisonAIAgents
+
+researcher = Agent(name="Researcher", role="Research Analyst")
+writer = Agent(name="Writer", role="Content Writer")
+
+research_task = Task(description="Research AI benefits", agent=researcher)
+write_task = Task(description="Write summary", agent=writer)
+
+agents = PraisonAIAgents(
+    agents=[researcher, writer],
+    tasks=[research_task, write_task],
+    planning=True,              # Enable planning
+    planning_tools=[search_web], # Tools for planning research
+    planning_reasoning=True,     # Chain-of-thought reasoning
+    auto_approve_plan=True       # Auto-approve plans
+)
+
+result = agents.start()
+```
+
+**Features:**
+- **Single Agent Planning**: `Agent(planning=True)` - simplest API
+- **Multi-Agent Planning**: `PraisonAIAgents(planning=True)` - orchestrated
+- **Planning Tools**: Research with tools before creating plans
+- **Reasoning Mode**: Chain-of-thought planning with tool use
+- **Todo Lists**: Auto-generated from plans with progress tracking
+- **Read-Only Mode**: `plan_mode=True` restricts agents to safe tools
+- **Approval Flow**: Review and approve plans before execution
+- **Persistence**: Plans saved to `.praison/plans/` as markdown
+
 ---
 
 ## Agent Types
@@ -216,27 +274,307 @@ router = RouterAgent(
 
 ## Memory
 
-Persistent memory with short-term, long-term, and graph memory support.
+PraisonAI provides a comprehensive memory system.
+
+### Quick Start (Zero Dependencies)
 
 ```python
-from praisonaiagents import Agent, Memory
+from praisonaiagents import Agent
+from praisonaiagents.memory import FileMemory
+
+# Simple file-based memory (no extra dependencies)
+memory = FileMemory(user_id="user123")
+
+agent = Agent(
+    name="Memory Agent",
+    instructions="You are a helpful assistant with memory.",
+    memory=memory  # or memory=True for auto FileMemory
+)
+
+response = agent.chat("My name is John and I prefer Python")
+# Memory automatically stores this context
+```
+
+### Advanced Memory (with providers)
+
+```python
+from praisonaiagents.memory import Memory
 
 memory = Memory(
-    provider="chroma",  # or "mongodb", "rag"
+    provider="chroma",  # or "mongodb", "mem0", "rag"
     use_short_term=True,
     use_long_term=True
 )
 
+agent = Agent(name="Memory Agent", memory=memory)
+```
+
+### Memory Types
+
+| Type | Description | Storage |
+|------|-------------|---------|
+| **Short-term** | Rolling buffer of recent context | JSON/SQLite |
+| **Long-term** | Persistent facts and knowledge | JSON/SQLite/Vector DB |
+| **Entity** | Named entities (people, places, concepts) | JSON/SQLite |
+| **Episodic** | Date-based interaction memories | JSON files |
+| **Graph** | Relationship-based memory | Mem0/Neo4j |
+
+### Session Save/Resume
+
+```python
+from praisonaiagents.memory import FileMemory
+
+memory = FileMemory(user_id="user123")
+
+# Add context during conversation
+memory.add_short_term("User is working on ML project")
+memory.add_long_term("User prefers Python", importance=0.9)
+
+# Save session for later
+memory.save_session("ml_project", conversation_history=[...])
+
+# Resume later
+session_data = memory.resume_session("ml_project")
+
+# List all sessions
+sessions = memory.list_sessions()
+```
+
+### Context Compression
+
+```python
+# Auto-compress when memory gets full
+memory.auto_compress_if_needed(threshold_percent=0.7)
+
+# Manual compression with LLM summarization
+def llm_summarize(prompt):
+    return agent.chat(prompt)
+
+summary = memory.compress(llm_func=llm_summarize, max_items=10)
+```
+
+### Checkpointing
+
+```python
+# Create checkpoint before risky operations
+checkpoint_id = memory.create_checkpoint("before_refactor", include_files=["main.py"])
+
+# Restore if needed
+memory.restore_checkpoint(checkpoint_id, restore_files=True)
+
+# List checkpoints
+checkpoints = memory.list_checkpoints()
+```
+
+### Memory Slash Commands
+
+```python
+# Handle slash commands programmatically
+result = memory.handle_command("/memory show")
+result = memory.handle_command("/memory add User likes coffee")
+result = memory.handle_command("/memory search Python")
+result = memory.handle_command("/memory save my_session")
+result = memory.handle_command("/memory compress")
+result = memory.handle_command("/memory checkpoint")
+result = memory.handle_command("/memory help")
+```
+
+**Available Commands:**
+- `/memory show` - Display stats and recent items
+- `/memory add <content>` - Add to long-term memory
+- `/memory clear [short|all]` - Clear memory
+- `/memory search <query>` - Search memories
+- `/memory save <name>` - Save session
+- `/memory resume <name>` - Resume session
+- `/memory sessions` - List saved sessions
+- `/memory compress` - Compress short-term memory
+- `/memory checkpoint [name]` - Create checkpoint
+- `/memory restore <id>` - Restore checkpoint
+- `/memory checkpoints` - List checkpoints
+- `/memory refresh` - Reload from disk
+
+---
+
+## Rules & Instructions
+
+PraisonAI automatically discovers and applies rules from multiple sources.
+
+### Supported Instruction Files
+
+| File | Description | Priority |
+|------|-------------|----------|
+| `PRAISON.md` | PraisonAI native instructions | High (500) |
+| `PRAISON.local.md` | Local overrides (gitignored) | Higher (600) |
+| `CLAUDE.md` | Claude Code memory file | High (500) |
+| `CLAUDE.local.md` | Local overrides (gitignored) | Higher (600) |
+| `AGENTS.md` | OpenAI Codex CLI instructions | High (500) |
+| `GEMINI.md` | Gemini CLI memory file | High (500) |
+| `.cursorrules` | Cursor IDE rules (legacy) | High (500) |
+| `.windsurfrules` | Windsurf IDE rules (legacy) | High (500) |
+| `.claude/rules/*.md` | Claude Code modular rules | Medium (50) |
+| `.windsurf/rules/*.md` | Windsurf modular rules | Medium (50) |
+| `.cursor/rules/*.mdc` | Cursor modular rules | Medium (50) |
+| `.praison/rules/*.md` | Workspace rules | Medium (0) |
+| `~/.praison/rules/*.md` | Global rules | Low (-1000) |
+
+### Auto-Discovery
+
+Rules are automatically loaded when you create an Agent:
+
+```python
+from praisonaiagents import Agent
+
+# Agent auto-discovers CLAUDE.md, AGENTS.md, GEMINI.md, etc.
 agent = Agent(
-    name="Memory Agent",
-    memory=memory
+    name="Assistant",
+    instructions="You are helpful."
+)
+
+# Rules are injected into system prompt automatically
+```
+
+### Rule File Format
+
+Rules support YAML frontmatter for advanced configuration:
+
+```markdown
+---
+description: Python coding guidelines
+globs: ["**/*.py", "**/*.pyx"]
+activation: always  # always, glob, manual, ai_decision
+priority: 10
+---
+
+# Python Guidelines
+- Use type hints for all functions
+- Follow PEP 8 style guide
+- Write docstrings for public APIs
+```
+
+### Activation Modes
+
+| Mode | Description |
+|------|-------------|
+| `always` | Always applied (default) |
+| `glob` | Applied when file matches glob pattern |
+| `manual` | Only when explicitly invoked via @mention |
+| `ai_decision` | AI decides when to apply |
+
+### Programmatic Rules Management
+
+```python
+from praisonaiagents.memory import RulesManager
+
+rules = RulesManager(workspace_path="/path/to/project")
+
+# Get all active rules
+active = rules.get_active_rules()
+
+# Get rules for specific file
+python_rules = rules.get_rules_for_file("src/main.py")
+
+# Build context for LLM
+context = rules.build_rules_context(file_path="src/main.py")
+
+# Create new rule
+rules.create_rule(
+    name="testing",
+    content="Always write tests first",
+    globs=["**/*.test.*"],
+    activation="glob"
+)
+
+# Get stats
+stats = rules.get_stats()
+# {'total_rules': 5, 'root_rules': 3, 'workspace_rules': 2, ...}
+```
+
+### Storage Structure
+
+```
+project/
+├── CLAUDE.md              # Auto-loaded (Claude Code)
+├── CLAUDE.local.md        # Local overrides (gitignored)
+├── AGENTS.md              # Auto-loaded (Codex CLI)
+├── GEMINI.md              # Auto-loaded (Gemini CLI)
+├── PRAISON.md             # Auto-loaded (PraisonAI)
+├── PRAISON.local.md       # Local overrides (gitignored)
+├── .claude/rules/         # Claude Code modular rules
+├── .windsurf/rules/       # Windsurf rules
+├── .cursor/rules/         # Cursor rules
+├── .praison/
+│   ├── rules/             # Workspace rules
+│   │   ├── python.md
+│   │   └── testing.md
+│   ├── workflows/         # Reusable workflows
+│   │   └── deploy.md
+│   ├── hooks.json         # Pre/post operation hooks
+│   └── memory/
+│       └── {user_id}/
+│           ├── short_term.json
+│           ├── long_term.json
+│           ├── sessions/
+│           └── checkpoints/
+└── ~/.praison/
+    └── rules/             # Global rules
+        └── global.md
+```
+
+### @Import Syntax
+
+Include other files in your rules:
+
+```markdown
+# CLAUDE.md
+See @README for project overview
+See @docs/architecture.md for system design
+@~/.praison/my-preferences.md
+```
+
+### Auto-Generated Memories
+
+```python
+from praisonaiagents.memory import FileMemory, AutoMemory
+
+memory = FileMemory(user_id="user123")
+auto = AutoMemory(memory, enabled=True)
+
+# Automatically extracts and stores memories from conversations
+memories = auto.process_interaction(
+    "My name is John and I prefer Python for backend work"
+)
+# Extracts: name="John", preference="Python for backend"
+```
+
+### Workflows
+
+Create reusable multi-step workflows:
+
+```python
+from praisonaiagents.memory import WorkflowManager
+
+manager = WorkflowManager()
+
+# Execute a workflow
+result = manager.execute(
+    "deploy",
+    executor=lambda prompt: agent.chat(prompt),
+    variables={"environment": "production"}
 )
 ```
 
-**Memory Types:**
-- **Short-term**: Recent conversation context
-- **Long-term**: Persistent storage across sessions
-- **Graph**: Relationship-based memory with Mem0
+### Hooks
+
+```python
+from praisonaiagents.memory import HooksManager
+
+hooks = HooksManager()
+
+# Register Python hooks
+hooks.register("pre_write_code", lambda ctx: print(f"Writing {ctx['file']}"))
+
+# Or configure in .praison/hooks.json
+```
 
 ---
 
@@ -317,12 +655,23 @@ agent = Agent(tools=[my_tool])
 
 ## MCP (Model Context Protocol)
 
-Connect to MCP servers for extended capabilities.
+Connect to MCP servers for extended capabilities. Supports **Protocol Revision 2025-11-25**.
+
+### MCP Transports
+
+| Transport | URL Pattern | Use Case |
+|-----------|-------------|----------|
+| **stdio** | Command string | Local NPX/Python servers |
+| **Streamable HTTP** | `http(s)://...` | Production deployments |
+| **WebSocket** | `ws://` or `wss://` | Real-time bidirectional |
+| **SSE (Legacy)** | `http(s)://.../sse` | Backward compatibility |
+
+### MCP Client (Consume MCP Servers)
 
 ```python
 from praisonaiagents import Agent, MCP
 
-# Local MCP server
+# stdio - Local MCP server
 agent = Agent(
     tools=MCP(
         command="npx",
@@ -330,9 +679,19 @@ agent = Agent(
     )
 )
 
-# Remote MCP server (SSE)
+# Streamable HTTP - Production server
 agent = Agent(
-    tools=MCP(url="http://localhost:8080/sse")
+    tools=MCP("https://api.example.com/mcp")
+)
+
+# WebSocket - Real-time bidirectional (NEW)
+agent = Agent(
+    tools=MCP("wss://api.example.com/mcp", auth_token="your-token")
+)
+
+# SSE (Legacy) - Backward compatibility
+agent = Agent(
+    tools=MCP("http://localhost:8080/sse")
 )
 
 # With environment variables
@@ -345,6 +704,193 @@ agent = Agent(
 )
 ```
 
+### MCP Features (Protocol 2025-11-25)
+
+- **Session Management**: Automatic `Mcp-Session-Id` handling
+- **Protocol Versioning**: `Mcp-Protocol-Version` header on all requests
+- **Resumability**: SSE stream recovery via `Last-Event-ID`
+- **Security**: Origin validation, DNS rebinding prevention
+- **WebSocket**: Auto-reconnect with exponential backoff
+
+### MCP Server (Expose Tools as MCP Server)
+
+Expose your Python functions as MCP tools for external clients (Claude Desktop, Cursor, etc.).
+
+```python
+from praisonaiagents.mcp import ToolsMCPServer
+
+# Define your tools as regular Python functions
+def search_web(query: str, max_results: int = 5) -> dict:
+    """Search the web for information."""
+    return {"results": [f"Result for {query}"]}
+
+def calculate(expression: str) -> dict:
+    """Evaluate a mathematical expression."""
+    return {"result": eval(expression)}
+
+# Create and run MCP server
+server = ToolsMCPServer(name="my-tools")
+server.register_tools([search_web, calculate])
+server.run()  # Starts stdio server (for Claude Desktop)
+
+# Or run as SSE server for web clients
+# server.run_sse(host="0.0.0.0", port=8080)
+```
+
+---
+
+## Native Web Search
+
+Enable real-time web search capabilities using LiteLLM's native web search.
+
+```python
+from praisonaiagents import Agent
+
+# Enable web search with a supported model
+agent = Agent(
+    name="Researcher",
+    instructions="You are a helpful research assistant.",
+    llm="openai/gpt-4o-search-preview",
+    web_search=True
+)
+
+result = agent.start("What are the latest developments in AI today?")
+```
+
+**Supported Providers:**
+- **OpenAI**: `gpt-4o-search-preview`, `gpt-4o-mini-search-preview`
+- **Anthropic**: `claude-3-5-sonnet-latest`, `claude-sonnet-4`
+- **Google/Gemini**: `gemini-2.0-flash`, `gemini-2.5-pro`
+- **xAI**: `grok-3`
+- **Perplexity**: All models
+
+---
+
+## Web Fetch
+
+Retrieve full content from specific URLs (Anthropic models only).
+
+```python
+from praisonaiagents import Agent
+
+# Enable web fetch with an Anthropic model
+agent = Agent(
+    name="Content Analyzer",
+    instructions="You can fetch and analyze web content.",
+    llm="anthropic/claude-sonnet-4-20250514",
+    web_fetch=True
+)
+
+result = agent.start("Fetch and summarize https://example.com")
+```
+
+**Web Fetch Options:**
+```python
+agent = Agent(
+    llm="anthropic/claude-sonnet-4-20250514",
+    web_fetch={
+        "max_uses": 10,
+        "allowed_domains": ["example.com"],
+        "citations": {"enabled": True}
+    }
+)
+```
+
+---
+
+## Prompt Caching
+
+Reduce costs and latency by caching parts of prompts.
+
+```python
+from praisonaiagents import Agent
+
+# Enable prompt caching for Anthropic models
+agent = Agent(
+    name="Legal Analyst",
+    instructions="You are an AI assistant analyzing legal documents." * 50,  # Long prompt
+    llm="anthropic/claude-3-5-sonnet-latest",
+    prompt_caching=True
+)
+
+# First call creates cache, subsequent calls use cached tokens
+result = agent.start("Summarize the key terms")
+```
+
+**Supported Providers:**
+- **OpenAI** (`openai/`) - Automatic caching for prompts ≥1024 tokens
+- **Anthropic** (`anthropic/`) - Manual caching with `cache_control`
+- **Bedrock** (`bedrock/`) - All models supporting prompt caching
+- **Deepseek** (`deepseek/`) - Works like OpenAI
+
+---
+
+## Claude Memory Tool (Beta)
+
+Enable Claude to store and retrieve information across conversations.
+
+```python
+from praisonaiagents import Agent
+
+# Enable Claude memory (Anthropic models only)
+agent = Agent(
+    name="Research Assistant",
+    llm="anthropic/claude-sonnet-4-20250514",
+    claude_memory=True
+)
+
+# Claude will automatically:
+# 1. Check /memories directory before tasks
+# 2. Store progress/learnings in files
+# 3. Reference memories in future conversations
+result = agent.start("Research AI trends and remember key findings")
+```
+
+**Features:**
+- File-based persistent memory in `.praison/claude_memory/`
+- Claude autonomously decides what to store/retrieve
+- Supports: view, create, str_replace, insert, delete, rename commands
+- Cross-conversation learning
+
+**Supported Models:** Claude Sonnet 4, Claude Opus 4, Claude Haiku 4.5
+
+---
+
+## Memory (Zero Dependencies)
+
+Enable persistent memory for agents without any extra packages.
+
+```python
+from praisonaiagents import Agent
+
+# Simple: Enable file-based memory (no extra dependencies!)
+agent = Agent(
+    name="Assistant",
+    memory=True  # Uses FileMemory by default
+)
+
+# Store memories
+agent.store_memory("User prefers dark mode", memory_type="short_term")
+agent.store_memory("User's name is John", memory_type="long_term", importance=0.9)
+
+# Get memory context for prompts
+context = agent.get_memory_context(query="What does the user prefer?")
+```
+
+**Memory Types:**
+- **Short-term**: Rolling buffer of recent context (auto-expires)
+- **Long-term**: Persistent important facts
+- **Entity**: People, places, organizations
+- **Episodic**: Date-based interaction history
+
+**Storage Providers:**
+| Provider | Dependencies | Use Case |
+|----------|-------------|----------|
+| `memory=True` or `"file"` | None | Default, zero-config |
+| `"sqlite"` | Built-in | Search, indexing |
+| `"chromadb"` | chromadb | Semantic/vector search |
+| `"mem0"` | mem0ai | Graph memory, cloud |
+
 ---
 
 ## Guardrails

src/praisonai-agents/examples/fast_context_example.py

@@ -0,0 +1,287 @@
+"""
+Fast Context Usage Examples.
+
+This file demonstrates how to use the Fast Context feature for rapid code search.
+
+Fast Context is inspired by Windsurf's SWE-grep approach:
+- Parallel tool execution (up to 8 concurrent calls)
+- Limited turns (max 4) for fast response
+- Lightweight model for search operations
+- Result caching for repeated queries
+"""
+
+# Example 1: Basic FastContext Usage
+# ==================================
+
+def example_basic_search():
+    """Basic search using FastContext."""
+    from praisonaiagents.context.fast import FastContext
+    
+    # Create FastContext instance
+    fc = FastContext(workspace_path=".")
+    
+    # Search for a pattern
+    result = fc.search("def authenticate")
+    
+    print(f"Found {result.total_files} files")
+    print(f"Search time: {result.search_time_ms}ms")
+    
+    # Print results
+    for file_match in result.files[:5]:
+        print(f"  - {file_match.path}")
+        for lr in file_match.line_ranges:
+            print(f"    Lines {lr.start}-{lr.end}")
+
+
+# Example 2: Search with Patterns
+# ===============================
+
+def example_search_with_patterns():
+    """Search with include/exclude patterns."""
+    from praisonaiagents.context.fast import FastContext
+    
+    fc = FastContext(workspace_path=".")
+    
+    # Search only in Python files
+    result = fc.search(
+        "class Agent",
+        include_patterns=["**/*.py"],
+        exclude_patterns=["**/tests/**", "**/test_*.py"]
+    )
+    
+    print(f"Found {result.total_files} files matching 'class Agent'")
+    for f in result.files:
+        print(f"  - {f.path}")
+
+
+# Example 3: Get Context for Agent
+# ================================
+
+def example_get_context_for_agent():
+    """Get formatted context string for an agent."""
+    from praisonaiagents.context.fast import FastContext
+    
+    fc = FastContext(workspace_path=".")
+    
+    # Get context formatted for injection into an agent
+    context = fc.get_context_for_agent(
+        "authentication handlers",
+        max_files=5,
+        max_lines_per_file=50
+    )
+    
+    print("Context for agent:")
+    print(context[:500] + "..." if len(context) > 500 else context)
+
+
+# Example 4: Agent with FastContext Integration
+# =============================================
+
+def example_agent_with_fast_context():
+    """Use FastContext with an Agent."""
+    from praisonaiagents import Agent
+    
+    # Create agent with FastContext enabled
+    agent = Agent(
+        name="CodeAssistant",
+        instructions="You are a helpful code assistant.",
+        fast_context=True,
+        fast_context_path=".",  # Current directory
+        fast_context_model="gpt-4o-mini",
+        fast_context_max_turns=4,
+        fast_context_parallelism=8
+    )
+    
+    # Use delegate_to_fast_context for code search
+    context = agent.delegate_to_fast_context("find authentication code")
+    
+    if context:
+        print("Found relevant code context:")
+        print(context[:300] + "...")
+    else:
+        print("No relevant code found")
+
+
+# Example 5: File Indexer
+# =======================
+
+def example_file_indexer():
+    """Use FileIndexer for fast file lookups."""
+    from praisonaiagents.context.fast.indexer import FileIndexer
+    
+    # Create and run indexer
+    indexer = FileIndexer(workspace_path=".")
+    count = indexer.index()
+    
+    print(f"Indexed {count} files")
+    
+    # Find files by pattern
+    py_files = indexer.find_by_pattern("**/*.py")
+    print(f"Found {len(py_files)} Python files")
+    
+    # Find files by extension
+    js_files = indexer.find_by_extension(".js")
+    print(f"Found {len(js_files)} JavaScript files")
+    
+    # Get stats
+    stats = indexer.get_stats()
+    print(f"Total size: {stats['total_size_mb']} MB")
+
+
+# Example 6: Symbol Indexer
+# =========================
+
+def example_symbol_indexer():
+    """Use SymbolIndexer to find code symbols."""
+    from praisonaiagents.context.fast.indexer import SymbolIndexer, SymbolType
+    
+    # Create and run indexer
+    indexer = SymbolIndexer(workspace_path=".")
+    count = indexer.index()
+    
+    print(f"Indexed {count} symbols")
+    
+    # Find symbols by name
+    agent_symbols = indexer.find_by_name("Agent")
+    print(f"Found {len(agent_symbols)} symbols containing 'Agent'")
+    
+    # Find all classes
+    classes = indexer.find_by_type(SymbolType.CLASS)
+    print(f"Found {len(classes)} classes")
+    
+    # Get stats
+    stats = indexer.get_stats()
+    print(f"Symbols by type: {stats['by_type']}")
+
+
+# Example 7: Context Injector
+# ===========================
+
+def example_context_injector():
+    """Use ContextInjector to format results for agents."""
+    from praisonaiagents.context.fast import FastContext
+    from praisonaiagents.context.fast.context_injector import (
+        ContextInjector,
+        InjectionConfig
+    )
+    
+    # Search for code
+    fc = FastContext(workspace_path=".")
+    result = fc.search("def main")
+    
+    # Configure injection
+    config = InjectionConfig(
+        max_tokens=2000,
+        max_files=5,
+        format_style="markdown",  # or "xml", "plain"
+        prioritize_precision=True
+    )
+    
+    # Inject into prompts
+    injector = ContextInjector(config)
+    formatted = injector.format_context(result)
+    
+    print("Formatted context:")
+    print(formatted[:500] + "..." if len(formatted) > 500 else formatted)
+
+
+# Example 8: Environment Variable Configuration
+# =============================================
+
+def example_env_config():
+    """Configure FastContext via environment variables."""
+    import os
+    
+    # Set environment variables (usually done in shell or .env file)
+    os.environ["FAST_CONTEXT_MODEL"] = "gpt-4o-mini"
+    os.environ["FAST_CONTEXT_MAX_TURNS"] = "4"
+    os.environ["FAST_CONTEXT_PARALLELISM"] = "8"
+    os.environ["FAST_CONTEXT_TIMEOUT"] = "30.0"
+    os.environ["FAST_CONTEXT_CACHE"] = "true"
+    os.environ["FAST_CONTEXT_CACHE_TTL"] = "300"
+    
+    from praisonaiagents.context.fast import FastContext
+    
+    # FastContext will use environment variables as defaults
+    fc = FastContext(workspace_path=".")
+    
+    print(f"Model: {fc.model}")
+    print(f"Max turns: {fc.max_turns}")
+    print(f"Max parallel: {fc.max_parallel}")
+    print(f"Cache enabled: {fc.cache_enabled}")
+
+
+# Example 9: Caching
+# ==================
+
+def example_caching():
+    """Demonstrate result caching."""
+    from praisonaiagents.context.fast import FastContext
+    
+    fc = FastContext(
+        workspace_path=".",
+        cache_enabled=True,
+        cache_ttl=300  # 5 minutes
+    )
+    
+    # First search - not cached
+    result1 = fc.search("def main")
+    print(f"First search: from_cache={result1.from_cache}, time={result1.search_time_ms}ms")
+    
+    # Second search - cached
+    result2 = fc.search("def main")
+    print(f"Second search: from_cache={result2.from_cache}, time={result2.search_time_ms}ms")
+    
+    # Clear cache
+    fc.clear_cache()
+    
+    # Third search - not cached
+    result3 = fc.search("def main")
+    print(f"Third search: from_cache={result3.from_cache}, time={result3.search_time_ms}ms")
+
+
+# Example 10: Context Manager
+# ===========================
+
+def example_context_manager():
+    """Use FastContext as a context manager."""
+    from praisonaiagents.context.fast import FastContext
+    
+    with FastContext(workspace_path=".") as fc:
+        result = fc.search("import")
+        print(f"Found {result.total_files} files with imports")
+    
+    # Resources are automatically cleaned up
+
+
+if __name__ == "__main__":
+    print("=" * 60)
+    print("Fast Context Examples")
+    print("=" * 60)
+    
+    print("\n1. Basic Search")
+    print("-" * 40)
+    example_basic_search()
+    
+    print("\n2. Search with Patterns")
+    print("-" * 40)
+    example_search_with_patterns()
+    
+    print("\n3. Get Context for Agent")
+    print("-" * 40)
+    example_get_context_for_agent()
+    
+    print("\n5. File Indexer")
+    print("-" * 40)
+    example_file_indexer()
+    
+    print("\n9. Caching")
+    print("-" * 40)
+    example_caching()
+    
+    print("\n10. Context Manager")
+    print("-" * 40)
+    example_context_manager()
+    
+    print("\n" + "=" * 60)
+    print("All examples completed!")

src/praisonai-agents/examples/mcp_server_example.py

@@ -0,0 +1,144 @@
+"""Example: Expose custom tools as MCP server.
+
+This example demonstrates how to create an MCP server that exposes
+custom Python functions as MCP tools. These tools can then be used
+by any MCP client (Claude Desktop, Cursor, etc.).
+
+Usage:
+    # Run as stdio server (for Claude Desktop, etc.)
+    python mcp_server_example.py
+    
+    # Run as SSE server (for web clients)
+    python mcp_server_example.py --sse --port 8080
+"""
+
+import argparse
+from typing import List, Dict, Any
+
+
+# Define your custom tools as regular Python functions
+# Type hints and docstrings are used to generate MCP schemas
+
+def search_web(query: str, max_results: int = 5) -> Dict[str, Any]:
+    """Search the web for information.
+    
+    Args:
+        query: The search query string
+        max_results: Maximum number of results to return (default: 5)
+    
+    Returns:
+        Search results with titles and URLs
+    """
+    # This is a mock implementation - replace with actual search logic
+    return {
+        "query": query,
+        "results": [
+            {"title": f"Result {i+1} for '{query}'", "url": f"https://example.com/{i+1}"}
+            for i in range(max_results)
+        ]
+    }
+
+
+def calculate(expression: str) -> Dict[str, Any]:
+    """Evaluate a mathematical expression.
+    
+    Args:
+        expression: A mathematical expression to evaluate (e.g., "2 + 2 * 3")
+    
+    Returns:
+        The result of the calculation
+    """
+    try:
+        # Safe evaluation of mathematical expressions
+        # In production, use a proper math parser
+        allowed_chars = set("0123456789+-*/.() ")
+        if not all(c in allowed_chars for c in expression):
+            return {"error": "Invalid characters in expression"}
+        
+        result = eval(expression)  # Note: Use safer alternative in production
+        return {"expression": expression, "result": result}
+    except Exception as e:
+        return {"error": str(e)}
+
+
+def get_weather(city: str, units: str = "celsius") -> Dict[str, Any]:
+    """Get current weather for a city.
+    
+    Args:
+        city: Name of the city
+        units: Temperature units - "celsius" or "fahrenheit" (default: celsius)
+    
+    Returns:
+        Weather information for the city
+    """
+    # Mock implementation - replace with actual weather API
+    return {
+        "city": city,
+        "temperature": 22 if units == "celsius" else 72,
+        "units": units,
+        "condition": "Sunny",
+        "humidity": 45
+    }
+
+
+def list_files(directory: str = ".", pattern: str = "*") -> List[str]:
+    """List files in a directory.
+    
+    Args:
+        directory: Directory path to list (default: current directory)
+        pattern: Glob pattern to filter files (default: *)
+    
+    Returns:
+        List of file names matching the pattern
+    """
+    import glob
+    import os
+    
+    try:
+        full_pattern = os.path.join(directory, pattern)
+        files = glob.glob(full_pattern)
+        return [os.path.basename(f) for f in files]
+    except Exception as e:
+        return [f"Error: {str(e)}"]
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Run PraisonAI Tools as MCP Server")
+    parser.add_argument("--sse", action="store_true", help="Use SSE transport instead of stdio")
+    parser.add_argument("--port", type=int, default=8080, help="Port for SSE server (default: 8080)")
+    parser.add_argument("--host", type=str, default="0.0.0.0", help="Host for SSE server")
+    parser.add_argument("--debug", action="store_true", help="Enable debug logging")
+    args = parser.parse_args()
+    
+    # Import MCP server components
+    from praisonaiagents.mcp import ToolsMCPServer
+    
+    # Create server with custom name
+    server = ToolsMCPServer(
+        name="praisonai-custom-tools",
+        debug=args.debug
+    )
+    
+    # Register all tools
+    server.register_tools([
+        search_web,
+        calculate,
+        get_weather,
+        list_files
+    ])
+    
+    # Print registered tools
+    print(f"📦 Registered {len(server.tools)} tools:")
+    for name in server.get_tool_names():
+        print(f"   - {name}")
+    print()
+    
+    # Run the server
+    if args.sse:
+        server.run_sse(host=args.host, port=args.port)
+    else:
+        server.run_stdio()
+
+
+if __name__ == "__main__":
+    main()

src/praisonai-agents/planning_example.py

@@ -0,0 +1,34 @@
+"""Simple Planning Mode Example: Plan first, then execute todo items."""
+
+from praisonaiagents import Agent
+from praisonaiagents.planning import PlanningAgent, TodoList
+
+# Step 1: Create agents
+researcher = Agent(name="Researcher", role="Research Analyst", verbose=True)
+writer = Agent(name="Writer", role="Content Writer", verbose=True)
+
+# Step 2: Create a plan using PlanningAgent
+planner = PlanningAgent(llm="gpt-4o-mini", verbose=1)
+plan = planner.create_plan_sync(
+    request="Write a short article about the top 3 benefits of meditation",
+    agents=[researcher, writer],
+    context="Keep it concise"
+)
+
+# Step 3: Create todo list from the plan
+todo = TodoList.from_plan(plan)
+
+# Step 4: Execute each todo item one by one
+for item in todo.items:
+    todo.start(item.id)
+    
+    # Pick the right agent based on the item's assigned agent
+    if item.agent == "Researcher":
+        agent = researcher
+    else:
+        agent = writer
+    
+    # Execute the task
+    result = agent.chat(item.description)
+    
+    todo.complete(item.id)

src/praisonai-agents/planning_simple.py

@@ -0,0 +1,21 @@
+"""Simplest Planning Mode Example - Just enable planning=True"""
+
+from praisonaiagents import Agent, Task, PraisonAIAgents
+
+# Create agents
+researcher = Agent(name="Researcher", role="Research Analyst")
+writer = Agent(name="Writer", role="Content Writer")
+
+# Create tasks
+task1 = Task(description="Research the benefits of meditation", agent=researcher)
+task2 = Task(description="Write a short article about meditation benefits", agent=writer)
+
+# Run with planning enabled - that's it!
+agents = PraisonAIAgents(
+    agents=[researcher, writer],
+    tasks=[task1, task2],
+    planning=True,
+    auto_approve_plan=True  # Auto-approve the generated plan
+)
+
+agents.start()

src/praisonai-agents/praisonaiagents/__init__.py

@@ -54,7 +54,57 @@ except ImportError:
     MCP = None
 from .session import Session
 from .memory.memory import Memory
+# Workflows - import from dedicated workflows module
+from .workflows import (
+    Workflow, WorkflowStep, WorkflowContext, StepResult,
+    Route, Parallel, Loop, Repeat,
+    route, parallel, loop, repeat,
+    Pipeline  # Alias for Workflow
+)
 from .guardrails import GuardrailResult, LLMGuardrail
+
+# Fast Context support (lazy loaded to avoid performance impact)
+def __getattr__(name):
+    """Lazy load FastContext to avoid impacting package load time."""
+    if name == "FastContext":
+        from praisonaiagents.context.fast import FastContext
+        return FastContext
+    elif name == "FastContextResult":
+        from praisonaiagents.context.fast import FastContextResult
+        return FastContextResult
+    elif name == "FileMatch":
+        from praisonaiagents.context.fast import FileMatch
+        return FileMatch
+    elif name == "LineRange":
+        from praisonaiagents.context.fast import LineRange
+        return LineRange
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+# Planning mode support
+try:
+    from .planning import (
+        Plan,
+        PlanStep,
+        TodoList,
+        TodoItem,
+        PlanStorage,
+        PlanningAgent,
+        ApprovalCallback,
+        READ_ONLY_TOOLS,
+        RESTRICTED_TOOLS,
+    )
+    _planning_available = True
+except ImportError:
+    _planning_available = False
+    Plan = None
+    PlanStep = None
+    TodoList = None
+    TodoItem = None
+    PlanStorage = None
+    PlanningAgent = None
+    ApprovalCallback = None
+    READ_ONLY_TOOLS = []
+    RESTRICTED_TOOLS = []
 from .agent.handoff import Handoff, handoff, handoff_filters, RECOMMENDED_PROMPT_PREFIX, prompt_with_handoff_instructions
 # Flow display
 try:
@@ -247,3 +297,11 @@ if _mcp_available:
 if FlowDisplay is not None:
     __all__.extend(['FlowDisplay', 'track_workflow'])
 
+# Add FastContext exports (lazy loaded)
+__all__.extend([
+    'FastContext',
+    'FastContextResult',
+    'FileMatch',
+    'LineRange'
+])
+

src/praisonai-agents/praisonaiagents/agent/agent.py

@@ -234,7 +234,23 @@ class Agent:
         max_guardrail_retries: int = 3,
         handoffs: Optional[List[Union['Agent', 'Handoff']]] = None,
         base_url: Optional[str] = None,
-        api_key: Optional[str] = None
+        api_key: Optional[str] = None,
+        web_search: Optional[Union[bool, Dict[str, Any]]] = None,
+        web_fetch: Optional[Union[bool, Dict[str, Any]]] = None,
+        prompt_caching: Optional[bool] = None,
+        claude_memory: Optional[Union[bool, Any]] = None,
+        plan_mode: bool = False,
+        planning: bool = False,
+        planning_tools: Optional[List[Any]] = None,
+        planning_reasoning: bool = False,
+        fast_context: bool = False,
+        fast_context_path: Optional[str] = None,
+        fast_context_model: str = "gpt-4o-mini",
+        fast_context_max_turns: int = 4,
+        fast_context_parallelism: int = 8,
+        fast_context_timeout: float = 30.0,
+        history_in_context: Optional[int] = None,
+        auto_save: Optional[str] = None
     ):
         """Initialize an Agent instance.
 
@@ -400,7 +416,11 @@ class Agent:
                         model=model_name,
                         base_url=base_url,
                         api_key=api_key,
-                        metrics=metrics
+                        metrics=metrics,
+                        web_search=web_search,
+                        web_fetch=web_fetch,
+                        prompt_caching=prompt_caching,
+                        claude_memory=claude_memory
                     )
                 self._using_custom_llm = True
             except ImportError as e:
@@ -435,6 +455,10 @@ class Agent:
                 if api_key:
                     llm_params['api_key'] = api_key
                 llm_params['metrics'] = metrics
+                llm_params['web_search'] = web_search
+                llm_params['web_fetch'] = web_fetch
+                llm_params['prompt_caching'] = prompt_caching
+                llm_params['claude_memory'] = claude_memory
                 self.llm_instance = LLM(**llm_params)
                 self._using_custom_llm = True
                 
@@ -471,7 +495,8 @@ class Agent:
         self.max_iter = max_iter
         self.max_rpm = max_rpm
         self.max_execution_time = max_execution_time
-        self.memory = memory
+        self._memory_instance = None
+        self._init_memory(memory, user_id)
         self.verbose = verbose
         self.allow_delegation = allow_delegation
         self.step_callback = step_callback
@@ -511,6 +536,36 @@ Your Goal: {self.goal}
         # Store user_id
         self.user_id = user_id or "praison"
         self.reasoning_steps = reasoning_steps
+        self.plan_mode = plan_mode  # Read-only mode for planning
+        self.planning = planning  # Enable planning mode
+        self.planning_tools = planning_tools  # Tools for planning phase
+        self.planning_reasoning = planning_reasoning  # Enable reasoning during planning
+        self._planning_agent = None  # Lazy loaded PlanningAgent
+        self.web_search = web_search
+        self.web_fetch = web_fetch
+        self.prompt_caching = prompt_caching
+        self.claude_memory = claude_memory
+        
+        # Session management
+        self.history_in_context = history_in_context  # Number of past sessions to include
+        self.auto_save = auto_save  # Session name for auto-saving
+        
+        # Initialize rules manager for persistent context (like Cursor/Windsurf)
+        self._rules_manager = None
+        self._init_rules_manager()
+        
+        # Handle web_search fallback: inject DuckDuckGo tool for unsupported models
+        if web_search and not self._model_supports_web_search():
+            from ..tools.duckduckgo_tools import internet_search
+            # Check if internet_search is not already in tools
+            tool_names = [getattr(t, '__name__', str(t)) for t in self.tools]
+            if 'internet_search' not in tool_names and 'duckduckgo' not in tool_names:
+                self.tools.append(internet_search)
+                logging.info("Model does not support native web search. Added DuckDuckGo fallback tool.")
+        
+        # Log warning if web_fetch is enabled but model doesn't support it
+        if web_fetch and not self._model_supports_web_fetch():
+            logging.warning(f"Model '{self.llm}' does not support native web fetch. Web fetch will be ignored.")
         
         # Initialize guardrail settings
         self.guardrail = guardrail
@@ -542,6 +597,15 @@ Your Goal: {self.goal}
             self._knowledge_config = knowledge_config
             self.knowledge = None  # Will be initialized on first use
 
+        # Fast Context configuration (lazy loaded)
+        self.fast_context_enabled = fast_context
+        self.fast_context_path = fast_context_path or os.getcwd()
+        self.fast_context_model = fast_context_model
+        self.fast_context_max_turns = fast_context_max_turns
+        self.fast_context_parallelism = fast_context_parallelism
+        self.fast_context_timeout = fast_context_timeout
+        self._fast_context_instance = None  # Lazy loaded
+
     @property
     def console(self):
         """Lazily initialize Rich Console only when needed."""
@@ -576,6 +640,345 @@ Your Goal: {self.goal}
             self._agent_id = str(uuid.uuid4())
         return self._agent_id
     
+    def get_available_tools(self) -> List[Any]:
+        """
+        Get tools available to this agent, filtered by plan_mode if enabled.
+        
+        In plan_mode, only read-only tools are available to prevent
+        modifications during the planning phase.
+        
+        Returns:
+            List of available tools
+        """
+        if not self.plan_mode:
+            return self.tools
+            
+        # Filter to read-only tools only
+        from ..planning import READ_ONLY_TOOLS, RESTRICTED_TOOLS
+        
+        filtered_tools = []
+        for tool in self.tools:
+            tool_name = getattr(tool, '__name__', str(tool)).lower()
+            
+            # Check if tool is in restricted list
+            is_restricted = any(
+                restricted.lower() in tool_name 
+                for restricted in RESTRICTED_TOOLS
+            )
+            
+            if not is_restricted:
+                filtered_tools.append(tool)
+                
+        return filtered_tools
+    
+    def _model_supports_web_search(self) -> bool:
+        """
+        Check if the agent's model supports native web search via LiteLLM.
+        
+        Returns:
+            bool: True if the model supports native web search, False otherwise
+        """
+        from ..llm.model_capabilities import supports_web_search
+        
+        # Get the model name
+        if hasattr(self, 'llm_instance') and self.llm_instance:
+            model_name = self.llm_instance.model
+        elif hasattr(self, 'llm') and self.llm:
+            model_name = self.llm
+        else:
+            model_name = "gpt-5-nano"
+        
+        return supports_web_search(model_name)
+    
+    def _model_supports_web_fetch(self) -> bool:
+        """
+        Check if the agent's model supports web fetch via LiteLLM.
+        
+        Web fetch allows the model to retrieve full content from specific URLs.
+        Currently only supported by Anthropic Claude models.
+        
+        Returns:
+            bool: True if the model supports web fetch, False otherwise
+        """
+        from ..llm.model_capabilities import supports_web_fetch
+        
+        # Get the model name
+        if hasattr(self, 'llm_instance') and self.llm_instance:
+            model_name = self.llm_instance.model
+        elif hasattr(self, 'llm') and self.llm:
+            model_name = self.llm
+        else:
+            model_name = "gpt-5-nano"
+        
+        return supports_web_fetch(model_name)
+    
+    def _model_supports_prompt_caching(self) -> bool:
+        """
+        Check if the agent's model supports prompt caching via LiteLLM.
+        
+        Prompt caching allows caching parts of prompts to reduce costs and latency.
+        Supported by OpenAI, Anthropic, Bedrock, and Deepseek.
+        
+        Returns:
+            bool: True if the model supports prompt caching, False otherwise
+        """
+        from ..llm.model_capabilities import supports_prompt_caching
+        
+        # Get the model name
+        if hasattr(self, 'llm_instance') and self.llm_instance:
+            model_name = self.llm_instance.model
+        elif hasattr(self, 'llm') and self.llm:
+            model_name = self.llm
+        else:
+            model_name = "gpt-4o-mini"
+        
+        return supports_prompt_caching(model_name)
+    
+    @property
+    def fast_context(self):
+        """Lazily initialize FastContext instance when needed.
+        
+        Returns:
+            FastContext instance or None if not enabled
+        """
+        if not self.fast_context_enabled:
+            return None
+        
+        if self._fast_context_instance is None:
+            try:
+                from ..context.fast import FastContext
+                self._fast_context_instance = FastContext(
+                    workspace_path=self.fast_context_path,
+                    model=self.fast_context_model,
+                    max_turns=self.fast_context_max_turns,
+                    max_parallel=self.fast_context_parallelism,
+                    timeout=self.fast_context_timeout,
+                    verbose=self.verbose
+                )
+            except ImportError:
+                logging.warning("FastContext not available")
+                return None
+        
+        return self._fast_context_instance
+    
+    def delegate_to_fast_context(self, query: str) -> Optional[str]:
+        """Delegate a code search query to FastContext subagent.
+        
+        This method uses the FastContext subagent to rapidly search
+        the codebase and return relevant context.
+        
+        Args:
+            query: Natural language search query
+            
+        Returns:
+            Formatted context string or None if FastContext not available
+        """
+        if not self.fast_context_enabled or self.fast_context is None:
+            return None
+        
+        try:
+            result = self.fast_context.search(query)
+            if result.total_files > 0:
+                return self.fast_context.get_context_for_agent(query)
+            return None
+        except Exception as e:
+            logging.warning(f"FastContext search failed: {e}")
+            return None
+    
+    def _init_rules_manager(self):
+        """
+        Initialize RulesManager for persistent rules/instructions.
+        
+        Automatically discovers rules from:
+        - ~/.praison/rules/ (global)
+        - .praison/rules/ (workspace)
+        - Subdirectory rules
+        """
+        try:
+            from ..memory.rules_manager import RulesManager
+            import os
+            
+            # Get workspace path (current working directory)
+            workspace_path = os.getcwd()
+            
+            self._rules_manager = RulesManager(
+                workspace_path=workspace_path,
+                verbose=1 if self.verbose else 0
+            )
+            
+            # Log discovered rules
+            stats = self._rules_manager.get_stats()
+            if stats["total_rules"] > 0:
+                logging.debug(f"RulesManager: Discovered {stats['total_rules']} rules")
+        except ImportError:
+            logging.debug("RulesManager not available")
+            self._rules_manager = None
+        except Exception as e:
+            logging.debug(f"Could not initialize RulesManager: {e}")
+            self._rules_manager = None
+    
+    def get_rules_context(self, file_path: Optional[str] = None, include_manual: Optional[List[str]] = None) -> str:
+        """
+        Get rules context for the current conversation.
+        
+        Args:
+            file_path: Optional file path for glob-based rule matching
+            include_manual: Optional list of manual rule names to include (via @mention)
+            
+        Returns:
+            Formatted rules context string
+        """
+        if not self._rules_manager:
+            return ""
+        
+        return self._rules_manager.build_rules_context(
+            file_path=file_path,
+            include_manual=include_manual
+        )
+    
+    def _init_memory(self, memory, user_id: Optional[str] = None):
+        """
+        Initialize memory based on the memory parameter.
+        
+        Args:
+            memory: Can be:
+                - True: Use FileMemory with default settings
+                - False/None: No memory
+                - "file": Use FileMemory
+                - "sqlite": Use existing Memory class with SQLite
+                - dict: Configuration for memory
+                - Memory/FileMemory instance: Use directly
+            user_id: User identifier for memory isolation
+        """
+        self.memory = memory
+        
+        if memory is None or memory is False:
+            self._memory_instance = None
+            return
+        
+        # Determine user_id
+        mem_user_id = user_id or getattr(self, 'user_id', None) or "default"
+        
+        if memory is True or memory == "file":
+            # Use FileMemory (zero dependencies)
+            from ..memory.file_memory import FileMemory
+            self._memory_instance = FileMemory(
+                user_id=mem_user_id,
+                verbose=1 if getattr(self, 'verbose', False) else 0
+            )
+        elif isinstance(memory, str) and memory in ("sqlite", "chromadb", "mem0", "mongodb"):
+            # Use full Memory class with specific provider
+            try:
+                from ..memory.memory import Memory
+                config = {"provider": memory if memory != "sqlite" else "rag"}
+                self._memory_instance = Memory(config)
+            except ImportError:
+                logging.warning(f"Memory provider '{memory}' requires additional dependencies. Falling back to FileMemory.")
+                from ..memory.file_memory import FileMemory
+                self._memory_instance = FileMemory(user_id=mem_user_id)
+        elif isinstance(memory, dict):
+            # Configuration dict
+            provider = memory.get("provider", "file")
+            if provider == "file":
+                from ..memory.file_memory import FileMemory
+                self._memory_instance = FileMemory(
+                    user_id=memory.get("user_id", mem_user_id),
+                    config=memory
+                )
+            else:
+                try:
+                    from ..memory.memory import Memory
+                    self._memory_instance = Memory(memory)
+                except ImportError:
+                    logging.warning("Full Memory class requires additional dependencies. Falling back to FileMemory.")
+                    from ..memory.file_memory import FileMemory
+                    self._memory_instance = FileMemory(user_id=mem_user_id)
+        else:
+            # Assume it's already a memory instance
+            self._memory_instance = memory
+    
+    def get_memory_context(self, query: Optional[str] = None) -> str:
+        """
+        Get memory context for the current conversation.
+        
+        Args:
+            query: Optional query to focus the context
+            
+        Returns:
+            Formatted memory context string
+        """
+        if not self._memory_instance:
+            return ""
+        
+        if hasattr(self._memory_instance, 'get_context'):
+            return self._memory_instance.get_context(query=query)
+        
+        return ""
+    
+    def store_memory(self, content: str, memory_type: str = "short_term", **kwargs):
+        """
+        Store content in memory.
+        
+        Args:
+            content: Content to store
+            memory_type: Type of memory (short_term, long_term, entity, episodic)
+            **kwargs: Additional arguments for the memory method
+        """
+        if not self._memory_instance:
+            return
+        
+        if memory_type == "short_term" and hasattr(self._memory_instance, 'add_short_term'):
+            self._memory_instance.add_short_term(content, **kwargs)
+        elif memory_type == "long_term" and hasattr(self._memory_instance, 'add_long_term'):
+            self._memory_instance.add_long_term(content, **kwargs)
+        elif memory_type == "entity" and hasattr(self._memory_instance, 'add_entity'):
+            self._memory_instance.add_entity(content, **kwargs)
+        elif memory_type == "episodic" and hasattr(self._memory_instance, 'add_episodic'):
+            self._memory_instance.add_episodic(content, **kwargs)
+    
+    def _display_memory_info(self):
+        """Display memory information to user in a friendly format."""
+        if not self._memory_instance:
+            return
+        
+        # Only display once per chat session
+        if hasattr(self, '_memory_displayed') and self._memory_displayed:
+            return
+        self._memory_displayed = True
+        
+        stats = self._memory_instance.get_stats()
+        short_count = stats.get('short_term_count', 0)
+        long_count = stats.get('long_term_count', 0)
+        entity_count = stats.get('entity_count', 0)
+        storage_path = stats.get('storage_path', '')
+        
+        total_memories = short_count + long_count + entity_count
+        
+        if total_memories > 0:
+            from rich.panel import Panel
+            from rich.text import Text
+            
+            # Build memory info text
+            info_parts = []
+            if long_count > 0:
+                info_parts.append(f"💾 {long_count} long-term")
+            if short_count > 0:
+                info_parts.append(f"⚡ {short_count} short-term")
+            if entity_count > 0:
+                info_parts.append(f"👤 {entity_count} entities")
+            
+            memory_text = Text()
+            memory_text.append("🧠 Memory loaded: ", style="bold cyan")
+            memory_text.append(" | ".join(info_parts))
+            memory_text.append(f"\n📁 Storage: {storage_path}", style="dim")
+            
+            self.console.print(Panel(
+                memory_text,
+                title="[bold]Agent Memory[/bold]",
+                border_style="cyan",
+                expand=False
+            ))
+    
     @property
     def llm_model(self):
         """Unified property to get the LLM model regardless of configuration type.
@@ -822,17 +1225,35 @@ Your Goal: {self.goal}
         if not self.use_system_prompt:
             return None
         
-        # Check cache first
-        tools_key = self._get_tools_cache_key(tools)
-        cache_key = f"{self.role}:{self.goal}:{tools_key}"
-        
-        if cache_key in self._system_prompt_cache:
-            return self._system_prompt_cache[cache_key]
+        # Check cache first (skip cache if memory is enabled since context is dynamic)
+        if not self._memory_instance:
+            tools_key = self._get_tools_cache_key(tools)
+            cache_key = f"{self.role}:{self.goal}:{tools_key}"
+            
+            if cache_key in self._system_prompt_cache:
+                return self._system_prompt_cache[cache_key]
+        else:
+            cache_key = None  # Don't cache when memory is enabled
             
         system_prompt = f"""{self.backstory}\n
 Your Role: {self.role}\n
 Your Goal: {self.goal}"""
         
+        # Add rules context if rules manager is enabled
+        if self._rules_manager:
+            rules_context = self.get_rules_context()
+            if rules_context:
+                system_prompt += f"\n\n## Rules (Guidelines you must follow)\n{rules_context}"
+        
+        # Add memory context if memory is enabled
+        if self._memory_instance:
+            memory_context = self.get_memory_context()
+            if memory_context:
+                system_prompt += f"\n\n## Memory (Information you remember about the user)\n{memory_context}"
+                # Display memory info to user if verbose
+                if self.verbose:
+                    self._display_memory_info()
+        
         # Add tool usage instructions if tools are available
         # Use provided tools or fall back to self.tools
         tools_to_use = tools if tools is not None else self.tools
@@ -862,9 +1283,9 @@ Your Goal: {self.goal}"""
             if tool_names:
                 system_prompt += f"\n\nYou have access to the following tools: {', '.join(tool_names)}. Use these tools when appropriate to help complete your tasks. Always use tools when they can help provide accurate information or perform actions."
         
-        # Cache the generated system prompt
+        # Cache the generated system prompt (only if cache_key is set, i.e., memory not enabled)
         # Simple cache size limit to prevent unbounded growth
-        if len(self._system_prompt_cache) < self._max_cache_size:
+        if cache_key and len(self._system_prompt_cache) < self._max_cache_size:
             self._system_prompt_cache[cache_key] = system_prompt
         return system_prompt
 
@@ -2150,21 +2571,189 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
 
     def run(self, prompt: str, **kwargs):
         """Alias for start() method"""
-        return self.start(prompt, **kwargs) 
+        return self.start(prompt, **kwargs)
+    
+    def _get_planning_agent(self):
+        """Lazy load PlanningAgent for planning mode."""
+        if self._planning_agent is None and self.planning:
+            from ..planning import PlanningAgent
+            self._planning_agent = PlanningAgent(
+                llm=self.llm if hasattr(self, 'llm') else (self.llm_instance.model if hasattr(self, 'llm_instance') else "gpt-4o-mini"),
+                tools=self.planning_tools,
+                reasoning=self.planning_reasoning,
+                verbose=1 if self.verbose else 0
+            )
+        return self._planning_agent
+    
+    def _start_with_planning(self, prompt: str, **kwargs):
+        """Execute with planning mode - creates plan then executes each step."""
+        from rich.console import Console
+        from rich.panel import Panel
+        from rich.markdown import Markdown
+        
+        console = Console()
+        
+        # Step 1: Create the plan
+        console.print("\n[bold blue]📋 PLANNING PHASE[/bold blue]")
+        console.print("[dim]Creating implementation plan...[/dim]\n")
+        
+        planner = self._get_planning_agent()
+        plan = planner.create_plan_sync(request=prompt, agents=[self])
+        
+        if not plan or not plan.steps:
+            console.print("[yellow]⚠️ Planning failed, falling back to direct execution[/yellow]")
+            return self.chat(prompt, **kwargs)
+        
+        # Display the plan
+        console.print(Panel(
+            Markdown(plan.to_markdown()),
+            title="[bold green]Generated Plan[/bold green]",
+            border_style="green"
+        ))
+        
+        # Step 2: Execute each step
+        console.print("\n[bold blue]🚀 EXECUTION PHASE[/bold blue]\n")
+        
+        results = []
+        context = ""
+        
+        for i, step in enumerate(plan.steps):
+            progress = (i + 1) / len(plan.steps)
+            bar_length = 30
+            filled = int(bar_length * progress)
+            bar = "█" * filled + "░" * (bar_length - filled)
+            
+            console.print(f"[dim]Progress: [{bar}] {progress * 100:.0f}%[/dim]")
+            console.print(f"\n[bold]📌 Step {i + 1}/{len(plan.steps)}:[/bold] {step.description[:60]}...")
+            
+            # Build prompt with context from previous steps
+            step_prompt = step.description
+            if context:
+                step_prompt = f"{step.description}\n\nContext from previous steps:\n{context}"
+            
+            # Execute the step
+            result = self.chat(step_prompt, **kwargs)
+            results.append({"step": i + 1, "description": step.description, "result": result})
+            
+            # Update context for next step (use full result, not truncated)
+            context += f"\n\nStep {i + 1} result: {result if result else 'No result'}"
+            
+            console.print(f"   [green]✅ Completed[/green]")
+        
+        console.print(f"\n[bold green]🎉 EXECUTION COMPLETE[/bold green]")
+        console.print(f"[dim]Progress: [{'█' * bar_length}] 100%[/dim]")
+        console.print(f"Completed {len(plan.steps)}/{len(plan.steps)} steps!\n")
+        
+        # Compile all results into a comprehensive final output
+        if len(results) > 1:
+            # Create a compilation prompt
+            all_results_text = "\n\n".join([
+                f"## Step {r['step']}: {r['description']}\n\n{r['result']}" 
+                for r in results
+            ])
+            
+            compilation_prompt = f"""You are tasked with compiling a comprehensive, detailed report from the following research steps.
+
+IMPORTANT: Write a DETAILED and COMPREHENSIVE report. Do NOT summarize or compress the information. 
+Include ALL relevant details, data, statistics, and findings from each step.
+Organize the information logically with clear sections and subsections.
+
+## Research Results to Compile:
+
+{all_results_text}
+
+## Instructions:
+1. Combine all the information into a single, well-organized document
+2. Preserve ALL details, numbers, statistics, and specific findings
+3. Use clear headings and subheadings
+4. Do not omit any important information
+5. Make it comprehensive and detailed
+
+Write the complete compiled report:"""
+            
+            console.print("\n[bold blue]📝 COMPILING FINAL REPORT[/bold blue]")
+            console.print("[dim]Creating comprehensive output from all steps...[/dim]\n")
+            
+            final_result = self.chat(compilation_prompt, **kwargs)
+            return final_result
+        
+        # Return the single result if only one step
+        return results[0]["result"] if results else None
 
     def start(self, prompt: str, **kwargs):
         """Start the agent with a prompt. This is a convenience method that wraps chat()."""
-        # Check if streaming is enabled (either from kwargs or agent's stream attribute)
-        stream_enabled = kwargs.get('stream', getattr(self, 'stream', False))
+        # Load history from past sessions if history_in_context is set
+        self._load_history_context()
         
-        if stream_enabled:
+        # Check if planning mode is enabled
+        if self.planning:
+            result = self._start_with_planning(prompt, **kwargs)
+        elif kwargs.get('stream', getattr(self, 'stream', False)):
             # Return a generator for streaming response
-            return self._start_stream(prompt, **kwargs)
+            result = self._start_stream(prompt, **kwargs)
         else:
             # Return regular chat response for backward compatibility
-            # Explicitly pass the resolved stream parameter to avoid chat() method default
-            kwargs['stream'] = stream_enabled
-            return self.chat(prompt, **kwargs)
+            kwargs['stream'] = False
+            result = self.chat(prompt, **kwargs)
+        
+        # Auto-save session if enabled
+        self._auto_save_session()
+        
+        return result
+    
+    def _load_history_context(self):
+        """Load history from past sessions into context if history_in_context is set."""
+        if not self.history_in_context or not self._memory_instance:
+            return
+        
+        try:
+            sessions = self._memory_instance.list_sessions()
+            if not sessions:
+                return
+            
+            # Get the last N sessions
+            sessions_to_load = sessions[:self.history_in_context]
+            
+            for session_info in reversed(sessions_to_load):
+                try:
+                    session_data = self._memory_instance.resume_session(session_info["name"])
+                    history = session_data.get("conversation_history", [])
+                    
+                    # Add history to chat_history with a marker
+                    for msg in history:
+                        if msg not in self.chat_history:
+                            # Mark as from history to avoid duplication
+                            msg_copy = msg.copy()
+                            msg_copy["_from_history"] = True
+                            self.chat_history.append(msg_copy)
+                except Exception:
+                    continue
+                    
+            if sessions_to_load:
+                logging.debug(f"Loaded history from {len(sessions_to_load)} past session(s)")
+        except Exception as e:
+            logging.debug(f"Error loading history context: {e}")
+    
+    def _auto_save_session(self):
+        """Auto-save session if auto_save is enabled."""
+        if not self.auto_save or not self._memory_instance:
+            return
+        
+        try:
+            # Filter out history markers before saving
+            clean_history = [
+                {k: v for k, v in msg.items() if k != "_from_history"}
+                for msg in self.chat_history
+            ]
+            
+            self._memory_instance.save_session(
+                name=self.auto_save,
+                conversation_history=clean_history,
+                metadata={"agent_name": self.name, "user_id": self.user_id}
+            )
+            logging.debug(f"Auto-saved session: {self.auto_save}")
+        except Exception as e:
+            logging.debug(f"Error auto-saving session: {e}")
 
     def _start_stream(self, prompt: str, **kwargs) -> Generator[str, None, None]:
         """Stream generator for real-time response chunks."""

src/praisonai-agents/praisonaiagents/agent/prompt_expander_agent.py

@@ -37,7 +37,7 @@ import logging
 from typing import List, Optional, Any, Dict
 from dataclasses import dataclass, field
 from enum import Enum
-
+from datetime import datetime
 
 class ExpandStrategy(Enum):
     """Enumeration of available prompt expansion strategies."""
@@ -88,6 +88,8 @@ class PromptExpanderAgent:
     PROMPTS = {
         "basic": """You are an expert at improving task prompts.
 
+Current date: {current_date}
+
 Given the following task prompt, expand it to be clearer and more actionable while preserving the original intent.
 
 Guidelines:
@@ -95,6 +97,7 @@ Guidelines:
 - Add clarity without changing the core request
 - Fix any ambiguity
 - Keep it concise but complete
+- Use the current date for any time-sensitive context
 
 Original prompt: {prompt}
 
@@ -102,6 +105,8 @@ Expanded prompt:""",
 
         "detailed": """You are an expert at creating comprehensive task prompts.
 
+Current date: {current_date}
+
 Given the following brief task prompt, expand it into a detailed, actionable prompt that provides rich context and clear guidance.
 
 Guidelines:
@@ -111,6 +116,7 @@ Guidelines:
 - Add relevant constraints or requirements
 - Include format guidance if applicable
 - Mention style or tone if relevant
+- Use the current date for any time-sensitive context
 
 Original prompt: {prompt}
 
@@ -118,6 +124,8 @@ Detailed expanded prompt:""",
 
         "structured": """You are an expert at creating well-structured task prompts.
 
+Current date: {current_date}
+
 Given the following task prompt, expand it into a structured format with clear sections.
 
 Guidelines:
@@ -129,6 +137,7 @@ Guidelines:
   * Style: Tone or style guidance
   * Constraints: Any limitations
 - Be specific and actionable
+- Use the current date for any time-sensitive context
 
 Original prompt: {prompt}
 
@@ -136,6 +145,8 @@ Structured expanded prompt:""",
 
         "creative": """You are an expert at crafting inspiring, creative task prompts.
 
+Current date: {current_date}
+
 Given the following task prompt, expand it with vivid language and creative direction while keeping it actionable.
 
 Guidelines:
@@ -145,6 +156,7 @@ Guidelines:
 - Add creative direction and artistic guidance
 - Make it exciting and engaging
 - Preserve the core task intent
+- Use the current date for any time-sensitive context
 
 Original prompt: {prompt}
 
@@ -152,6 +164,8 @@ Creative expanded prompt:""",
 
         "auto": """You are an expert at analyzing and expanding task prompts.
 
+Current date: {current_date}
+
 Analyze the following prompt and expand it appropriately based on its nature.
 
 For short/vague prompts: Add significant detail and context
@@ -160,6 +174,7 @@ For creative tasks: Add artistic direction
 For technical tasks: Add precision and requirements
 
 IMPORTANT: Keep it as a TASK/COMMAND. Never convert to a question.
+Use the current date for any time-sensitive context.
 
 Original prompt: {prompt}
 {context}
@@ -243,11 +258,24 @@ Expanded prompt:"""
         prompt_lower = prompt.lower().strip()
         return any(prompt_lower.startswith(verb) for verb in task_verbs)
     
+    def _is_technical_task(self, prompt: str) -> bool:
+        """Check if prompt is a technical/coding task."""
+        technical_indicators = [
+            "python", "javascript", "java", "code", "function", "class",
+            "api", "database", "sql", "html", "css", "react", "node",
+            "algorithm", "debug", "fix", "error", "bug", "test",
+            "deploy", "server", "docker", "kubernetes", "aws", "azure",
+            "typescript", "rust", "go", "c++", "c#", "ruby", "php",
+            "script", "bash", "shell", "terminal", "command", "cli"
+        ]
+        prompt_lower = prompt.lower()
+        return any(ind in prompt_lower for ind in technical_indicators)
+    
     def _is_creative_task(self, prompt: str) -> bool:
-        """Check if prompt is a creative task."""
+        """Check if prompt is a creative task (non-technical)."""
         creative_indicators = [
-            "poem", "story", "script", "song", "art", "creative",
-            "fiction", "novel", "essay", "blog", "article", "write"
+            "poem", "story", "song", "art", "creative",
+            "fiction", "novel", "essay", "blog", "article"
         ]
         prompt_lower = prompt.lower()
         return any(ind in prompt_lower for ind in creative_indicators)
@@ -258,6 +286,10 @@ Expanded prompt:"""
         if self._is_short_prompt(prompt, threshold=5):
             return ExpandStrategy.DETAILED
         
+        # Technical/coding tasks get structured expansion (check before creative)
+        if self._is_technical_task(prompt):
+            return ExpandStrategy.STRUCTURED
+        
         # Creative tasks benefit from creative expansion
         if self._is_creative_task(prompt):
             return ExpandStrategy.CREATIVE
@@ -306,11 +338,46 @@ Expanded prompt:"""
             # Fallback if auto-detect returns AUTO
             strategy = ExpandStrategy.DETAILED
         
+        # Get current date for time-sensitive context
+        current_date = datetime.now().strftime("%B %d, %Y")
+        
         expansion_prompt = self.PROMPTS[strategy.value].format(
             prompt=prompt,
-            context=context or ""
+            context=context or "",
+            current_date=current_date
         )
         
+        # Add tool usage instructions if tools are available
+        if self.tools:
+            # Build dynamic tool descriptions from docstrings
+            tool_descriptions = []
+            for tool in self.tools:
+                tool_name = getattr(tool, '__name__', str(tool))
+                tool_doc = getattr(tool, '__doc__', '')
+                # Get first line of docstring as description
+                if tool_doc:
+                    first_line = tool_doc.strip().split('\n')[0].strip()
+                    tool_descriptions.append(f"- {tool_name}: {first_line}")
+                else:
+                    tool_descriptions.append(f"- {tool_name}")
+            
+            tools_list = '\n'.join(tool_descriptions)
+            
+            tool_instruction = f"""
+
+IMPORTANT: You have the following tools available:
+{tools_list}
+
+Before expanding this prompt, FIRST use the available tools to gather relevant current information about the topic. This will help you create a more informed and accurate expanded prompt.
+
+Steps:
+1. Use the appropriate tool(s) to gather relevant information about the topic
+2. Use the gathered information to inform your expansion
+3. Create an expanded prompt that incorporates the latest context
+
+"""
+            expansion_prompt = tool_instruction + expansion_prompt
+        
         # Call the agent
         expanded = self._call_agent(expansion_prompt).strip()
         

src/praisonai-agents/praisonaiagents/agents/agents.py

@@ -2,7 +2,7 @@ import os
 import time
 import json
 import logging
-from typing import Any, Dict, Optional, List
+from typing import Any, Dict, Optional, List, Callable
 from pydantic import BaseModel
 from rich.text import Text
 from rich.panel import Panel
@@ -127,7 +127,7 @@ def process_task_context(context_item, verbose=0, user_id=None):
         return str(context_item)  # Fallback for unknown types
 
 class PraisonAIAgents:
-    def __init__(self, agents, tasks=None, verbose=0, completion_checker=None, max_retries=5, process="sequential", manager_llm=None, memory=False, memory_config=None, embedder=None, user_id=None, max_iter=10, stream=True, name: Optional[str] = None):
+    def __init__(self, agents, tasks=None, verbose=0, completion_checker=None, max_retries=5, process="sequential", manager_llm=None, memory=False, memory_config=None, embedder=None, user_id=None, max_iter=10, stream=True, name: Optional[str] = None, planning: bool = False, planning_llm: Optional[str] = None, auto_approve_plan: bool = False, planning_tools: Optional[List] = None, planning_reasoning: bool = False, on_task_start: Optional[Callable] = None, on_task_complete: Optional[Callable] = None, variables: Optional[Dict[str, Any]] = None):
         # Add check at the start if memory is requested
         if memory:
             try:
@@ -162,6 +162,11 @@ class PraisonAIAgents:
         self.stream = stream
         self.name = name  # Store the name for the Agents collection
         
+        # Callbacks for workflow execution
+        self.on_task_start = on_task_start  # Called before each task
+        self.on_task_complete = on_task_complete  # Called after each task
+        self.variables = variables if variables else {}  # Global variables for substitution
+        
         # Check for manager_llm in environment variable if not provided
         self.manager_llm = manager_llm or os.getenv('OPENAI_MODEL_NAME', 'gpt-5-nano')
         
@@ -213,6 +218,16 @@ class PraisonAIAgents:
         
         self._state = {}  # Add state storage at PraisonAIAgents level
         
+        # Initialize planning mode
+        self.planning = planning
+        self.planning_llm = planning_llm or "gpt-4o-mini"
+        self.auto_approve_plan = auto_approve_plan
+        self.planning_tools = planning_tools
+        self.planning_reasoning = planning_reasoning
+        self._current_plan = None
+        self._todo_list = None
+        self._planning_agent = None
+        
         # Initialize memory system
         self.shared_memory = None
         if memory:
@@ -651,13 +666,24 @@ Context:
 
         executor_agent = task.agent
         
+        # Create agent from agent_config if provided and no agent assigned
+        if executor_agent is None and getattr(task, 'agent_config', None):
+            executor_agent = self._create_agent_from_config(task.agent_config)
+            task.agent = executor_agent
+        
         # Set current agent for token tracking
         llm = getattr(executor_agent, 'llm', None) or getattr(executor_agent, 'llm_instance', None)
         if llm and hasattr(llm, 'set_current_agent'):
             llm.set_current_agent(executor_agent.name)
 
+        # Substitute variables in task description if provided
+        task_description = task.description
+        if getattr(task, 'variables', None):
+            for key, value in task.variables.items():
+                task_description = task_description.replace(f"{{{{{key}}}}}", str(value))
+
         task_prompt = f"""
-You need to do the following task: {task.description}.
+You need to do the following task: {task_description}.
 Expected Output: {task.expected_output}.
 """
         if task.context:
@@ -815,6 +841,17 @@ Context:
             logger.info(f"Task with ID {task_id} is already completed")
             return
 
+        # Call on_task_start callback if provided
+        if self.on_task_start:
+            try:
+                self.on_task_start(task, task_id)
+            except Exception as e:
+                logger.error(f"Error in on_task_start callback: {e}")
+        
+        # Apply global variables to task if not already set
+        if self.variables and not getattr(task, 'variables', None):
+            task.variables = self.variables
+        
         retries = 0
         while task.status != "completed" and retries < self.max_retries:
             logger.debug(f"Attempt {retries+1} for task {task_id}")
@@ -847,6 +884,14 @@ Context:
                             logger.exception(e)
                             
                     self.save_output_to_file(task, task_output)
+                    
+                    # Call on_task_complete callback if provided
+                    if self.on_task_complete:
+                        try:
+                            self.on_task_complete(task, task_output)
+                        except Exception as e:
+                            logger.error(f"Error in on_task_complete callback: {e}")
+                    
                     if self.verbose >= 1:
                         logger.info(f"Task {task_id} completed successfully.")
                 else:
@@ -929,9 +974,13 @@ Context:
                         task.context = []
                     # Add content to context
                     task.context.append(content)
-                
-        # Run tasks as before
-        self.run_all_tasks()
+        
+        # Planning Mode: Create plan and todo list before execution
+        if self.planning:
+            self._run_with_planning()
+        else:
+            # Run tasks as before
+            self.run_all_tasks()
         
         # Auto-display token metrics if any agent has metrics=True
         metrics_enabled = any(getattr(agent, 'metrics', False) for agent in self.agents)
@@ -1532,3 +1581,377 @@ Context:
         else:
             display_error(f"Invalid protocol: {protocol}. Choose 'http' or 'mcp'.")
             return None
+
+    # =========================================================================
+    # Planning Mode Properties and Methods
+    # =========================================================================
+    
+    @property
+    def current_plan(self):
+        """Get the current plan."""
+        return self._current_plan
+    
+    @property
+    def todo_list(self):
+        """Get the current todo list."""
+        return self._todo_list
+    
+    def _create_agent_from_config(self, agent_config: dict) -> 'Agent':
+        """
+        Create an Agent from a configuration dictionary.
+        
+        Args:
+            agent_config: Dict with keys like 'name', 'role', 'goal', 'backstory', 'llm', 'tools'
+            
+        Returns:
+            Agent instance
+        """
+        from ..agent.agent import Agent
+        
+        return Agent(
+            name=agent_config.get('name', 'TaskAgent'),
+            role=agent_config.get('role'),
+            goal=agent_config.get('goal'),
+            backstory=agent_config.get('backstory'),
+            llm=agent_config.get('llm'),
+            tools=agent_config.get('tools'),
+            verbose=self.verbose >= 1,
+            memory=self.memory
+        )
+    
+    def _get_planning_agent(self):
+        """Lazy load PlanningAgent."""
+        if self._planning_agent is None and self.planning:
+            from ..planning import PlanningAgent
+            self._planning_agent = PlanningAgent(
+                llm=self.planning_llm,
+                read_only=True,
+                verbose=self.verbose,
+                tools=self.planning_tools,
+                reasoning=self.planning_reasoning
+            )
+        return self._planning_agent
+    
+    async def _create_plan(self, request: str = None, context: str = None):
+        """
+        Create an implementation plan.
+        
+        Args:
+            request: The request/goal to plan for
+            context: Optional additional context
+            
+        Returns:
+            Plan instance
+        """
+        planner = self._get_planning_agent()
+        if planner is None:
+            return None
+            
+        # Use first task description as request if not provided
+        if request is None and self.tasks:
+            first_task = list(self.tasks.values())[0]
+            request = first_task.description
+            
+        plan = await planner.create_plan(
+            request=request,
+            agents=self.agents,
+            tasks=list(self.tasks.values()),
+            context=context
+        )
+        
+        self._current_plan = plan
+        
+        # Create todo list from plan
+        from ..planning import TodoList
+        self._todo_list = TodoList.from_plan(plan)
+        
+        return plan
+    
+    def _create_plan_sync(self, request: str = None, context: str = None):
+        """
+        Synchronous version of _create_plan.
+        
+        Args:
+            request: The request/goal to plan for
+            context: Optional additional context
+            
+        Returns:
+            Plan instance
+        """
+        planner = self._get_planning_agent()
+        if planner is None:
+            return None
+            
+        # Use first task description as request if not provided
+        if request is None and self.tasks:
+            first_task = list(self.tasks.values())[0]
+            request = first_task.description
+            
+        plan = planner.create_plan_sync(
+            request=request,
+            agents=self.agents,
+            tasks=list(self.tasks.values()),
+            context=context
+        )
+        
+        self._current_plan = plan
+        
+        # Create todo list from plan
+        from ..planning import TodoList
+        self._todo_list = TodoList.from_plan(plan)
+        
+        return plan
+    
+    async def _request_approval(self, plan):
+        """
+        Request approval for a plan.
+        
+        Args:
+            plan: Plan to approve
+            
+        Returns:
+            True if approved, False if rejected
+        """
+        if self.auto_approve_plan:
+            plan.approve()
+            return True
+            
+        from ..planning import ApprovalCallback
+        callback = ApprovalCallback(auto_approve=self.auto_approve_plan)
+        return await callback.async_call(plan)
+    
+    def _request_approval_sync(self, plan):
+        """
+        Synchronous version of _request_approval.
+        
+        Args:
+            plan: Plan to approve
+            
+        Returns:
+            True if approved, False if rejected
+        """
+        if self.auto_approve_plan:
+            plan.approve()
+            return True
+            
+        from ..planning import ApprovalCallback
+        callback = ApprovalCallback(auto_approve=self.auto_approve_plan)
+        return callback(plan)
+    
+    def get_plan_markdown(self) -> str:
+        """
+        Get the current plan as markdown.
+        
+        Returns:
+            Markdown string or empty string if no plan
+        """
+        if self._current_plan:
+            return self._current_plan.to_markdown()
+        return ""
+    
+    def get_todo_markdown(self) -> str:
+        """
+        Get the current todo list as markdown.
+        
+        Returns:
+            Markdown string or empty string if no todo list
+        """
+        if self._todo_list:
+            return self._todo_list.to_markdown()
+        return ""
+    
+    def update_plan_step_status(self, step_id: str, status: str) -> bool:
+        """
+        Update the status of a plan step.
+        
+        Args:
+            step_id: ID of the step to update
+            status: New status
+            
+        Returns:
+            True if updated, False if not found
+        """
+        if self._current_plan:
+            result = self._current_plan.update_step_status(step_id, status)
+            # Sync todo list
+            if self._todo_list:
+                self._todo_list.sync_with_plan(self._current_plan)
+            return result
+        return False
+    
+    def _run_with_planning(self):
+        """
+        Run tasks with planning mode enabled.
+        
+        This method:
+        1. Creates a plan using PlanningAgent
+        2. Generates a todo list from the plan
+        3. Creates proper Task objects from plan steps
+        4. Executes each task using the full Task execution system
+           (with memory, callbacks, guardrails, structured output, etc.)
+        5. Tracks progress as items are completed
+        """
+        from rich.console import Console
+        from rich.panel import Panel
+        from rich.markdown import Markdown
+        from ..task import Task
+        
+        console = Console()
+        
+        # Step 1: Create the plan
+        console.print("\n[bold blue]📋 PLANNING PHASE[/bold blue]")
+        console.print("[dim]Creating implementation plan...[/dim]\n")
+        
+        # Build request from tasks
+        task_descriptions = [task.description for task in self.tasks.values()]
+        request = " AND ".join(task_descriptions)
+        
+        plan = self._create_plan_sync(request=request)
+        
+        if not plan:
+            console.print("[yellow]⚠️ Planning failed, falling back to normal execution[/yellow]")
+            self.run_all_tasks()
+            return
+        
+        # Display the plan
+        console.print(Panel(
+            Markdown(plan.to_markdown()),
+            title="[bold green]Generated Plan[/bold green]",
+            border_style="green"
+        ))
+        
+        # Step 2: Request approval if not auto-approve
+        if not self.auto_approve_plan:
+            approved = self._request_approval_sync(plan)
+            if not approved:
+                console.print("[red]❌ Plan rejected. Aborting execution.[/red]")
+                return
+        else:
+            plan.approve()
+            console.print("[green]✅ Plan auto-approved[/green]")
+        
+        # Step 3: Create todo list for tracking
+        from ..planning import TodoList
+        self._todo_list = TodoList.from_plan(plan)
+        
+        console.print("\n[bold blue]📝 TODO LIST[/bold blue]")
+        console.print(Panel(
+            Markdown(self._todo_list.to_markdown()),
+            title="[bold cyan]Tasks to Complete[/bold cyan]",
+            border_style="cyan"
+        ))
+        
+        # Step 4: Create proper Task objects from plan steps
+        console.print("\n[bold blue]🚀 EXECUTION PHASE[/bold blue]\n")
+        
+        # Map agent names to agent instances
+        agent_map = {agent.name: agent for agent in self.agents}
+        
+        # Store original tasks and create new tasks from plan
+        original_tasks = self.tasks.copy()
+        self.tasks = {}
+        self.task_id_counter = 0
+        
+        # Create Task objects from plan steps
+        plan_tasks = []
+        step_to_task = {}  # Map step_id to task for context chaining
+        
+        for i, step in enumerate(plan.steps):
+            # Get the appropriate agent
+            agent = agent_map.get(step.agent, self.agents[0] if self.agents else None)
+            
+            if not agent:
+                console.print(f"[yellow]⚠️ No agent found for '{step.agent}', using first available[/yellow]")
+                agent = self.agents[0] if self.agents else None
+            
+            if not agent:
+                console.print(f"[red]❌ No agents available for step: {step.description}[/red]")
+                continue
+            
+            # Build context from dependencies (previous task results)
+            context = []
+            for dep_id in step.dependencies:
+                # Convert step_X format to actual step index
+                if dep_id.startswith("step_"):
+                    try:
+                        dep_index = int(dep_id.split("_")[1])
+                        if dep_index < len(plan_tasks):
+                            context.append(plan_tasks[dep_index])
+                    except (ValueError, IndexError):
+                        pass
+                elif dep_id in step_to_task:
+                    context.append(step_to_task[dep_id])
+            
+            # Find matching original task for additional config (memory, callbacks, etc.)
+            original_task = None
+            for orig_task in original_tasks.values():
+                if orig_task.agent and orig_task.agent.name == agent.name:
+                    original_task = orig_task
+                    break
+            
+            # Create Task with full features from original task if available
+            task = Task(
+                description=step.description,
+                expected_output=f"Complete: {step.description}",
+                agent=agent,
+                name=f"Plan Step {i + 1}",
+                tools=agent.tools if agent.tools else [],
+                context=context if context else None,
+                # Inherit from original task if available
+                memory=original_task.memory if original_task else None,
+                callback=original_task.callback if original_task else None,
+                guardrail=original_task.guardrail if original_task else None,
+                max_retries=original_task.max_retries if original_task else 3,
+                output_json=original_task.output_json if original_task else None,
+                output_pydantic=original_task.output_pydantic if original_task else None,
+                config=original_task.config if original_task else {}
+            )
+            
+            # Add task to our task list
+            task_id = self.add_task(task)
+            plan_tasks.append(task)
+            step_to_task[step.id] = task
+        
+        # Step 5: Execute tasks using the proper Task execution system
+        for i, (task_id, task) in enumerate(self.tasks.items()):
+            # Update todo list progress
+            if i < len(self._todo_list.items):
+                item = self._todo_list.items[i]
+                
+                # Display progress bar
+                progress = self._todo_list.progress
+                bar_length = 30
+                filled = int(bar_length * progress)
+                bar = "█" * filled + "░" * (bar_length - filled)
+                console.print(f"[dim]Progress: [{bar}] {progress * 100:.0f}%[/dim]")
+                
+                console.print(f"\n[bold]📌 Step {i + 1}/{len(self.tasks)}:[/bold] {task.description[:60]}...")
+                console.print(f"[dim]   Agent: {task.agent.name if task.agent else 'Unknown'}[/dim]")
+                
+                # Mark as in progress
+                self._todo_list.start(item.id)
+            
+            # Execute using the full Task execution system
+            # This includes: memory, callbacks, guardrails, structured output, retry logic
+            try:
+                self.run_task(task_id)
+                
+                if task.status == "completed":
+                    if i < len(self._todo_list.items):
+                        self._todo_list.complete(self._todo_list.items[i].id)
+                    console.print("[green]   ✅ Completed[/green]")
+                else:
+                    console.print(f"[yellow]   ⚠️ Task status: {task.status}[/yellow]")
+            except Exception as e:
+                console.print(f"[red]   ❌ Error: {e}[/red]")
+                logger.error(f"Error executing plan task {task_id}: {e}")
+        
+        # Final progress
+        completed_count = len([t for t in self.tasks.values() if t.status == "completed"])
+        console.print(f"\n[bold green]🎉 EXECUTION COMPLETE[/bold green]")
+        console.print(f"[dim]Progress: [{'█' * 30}] 100%[/dim]")
+        console.print(f"[green]Completed {completed_count}/{len(self.tasks)} tasks![/green]\n")
+        
+        # Restore original tasks reference for result retrieval
+        self._plan_tasks = self.tasks.copy()
+        # Keep plan tasks for results but note original tasks are preserved in _plan_tasks

src/praisonai-agents/praisonaiagents/context/__init__.py

@@ -0,0 +1,30 @@
+"""
+Context module for PraisonAI Agents.
+
+This module provides context retrieval capabilities including:
+- Fast Context: Rapid parallel code search subagent
+"""
+
+# Lazy imports to avoid impacting package load time
+def __getattr__(name):
+    """Lazy import to avoid impacting package performance."""
+    if name == "FastContext":
+        from praisonaiagents.context.fast.fast_context import FastContext
+        return FastContext
+    elif name == "FastContextResult":
+        from praisonaiagents.context.fast.result import FastContextResult
+        return FastContextResult
+    elif name == "FileMatch":
+        from praisonaiagents.context.fast.result import FileMatch
+        return FileMatch
+    elif name == "LineRange":
+        from praisonaiagents.context.fast.result import LineRange
+        return LineRange
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+__all__ = [
+    "FastContext",
+    "FastContextResult",
+    "FileMatch",
+    "LineRange",
+]

src/praisonai-agents/praisonaiagents/context/fast/__init__.py

@@ -0,0 +1,40 @@
+"""
+Fast Context module for rapid parallel code search.
+
+This module implements a specialized subagent that retrieves relevant code
+from a codebase using parallel tool calls, similar to Windsurf's Fast Context.
+
+Key features:
+- Parallel tool execution (up to 8 concurrent calls)
+- Limited serial turns (max 4) for fast response
+- Restricted tool set (grep, glob, read) for safety
+- Returns files + line ranges (no summarization)
+"""
+
+# Lazy imports to avoid impacting package load time
+def __getattr__(name):
+    """Lazy import to avoid impacting package performance."""
+    if name == "FastContext":
+        from praisonaiagents.context.fast.fast_context import FastContext
+        return FastContext
+    elif name == "FastContextResult":
+        from praisonaiagents.context.fast.result import FastContextResult
+        return FastContextResult
+    elif name == "FileMatch":
+        from praisonaiagents.context.fast.result import FileMatch
+        return FileMatch
+    elif name == "LineRange":
+        from praisonaiagents.context.fast.result import LineRange
+        return LineRange
+    elif name == "FastContextAgent":
+        from praisonaiagents.context.fast.fast_context_agent import FastContextAgent
+        return FastContextAgent
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+__all__ = [
+    "FastContext",
+    "FastContextResult",
+    "FileMatch",
+    "LineRange",
+    "FastContextAgent",
+]

src/praisonai-agents/praisonaiagents/context/fast/context_injector.py

@@ -0,0 +1,383 @@
+"""
+Context Injector for Fast Context.
+
+Injects FastContext results into the main agent with:
+- Token budget management
+- Context formatting for different models
+- Precision over recall (avoid context pollution)
+"""
+
+import logging
+from typing import Optional, List, Dict, Any
+from dataclasses import dataclass
+
+from praisonaiagents.context.fast.result import FastContextResult, FileMatch
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class InjectionConfig:
+    """Configuration for context injection.
+    
+    Attributes:
+        max_tokens: Maximum tokens for injected context
+        max_files: Maximum files to include
+        max_lines_per_file: Maximum lines per file
+        include_line_numbers: Whether to include line numbers
+        include_file_content: Whether to include file content
+        format_style: Format style ('markdown', 'xml', 'plain')
+        prioritize_precision: If True, include fewer but more relevant results
+    """
+    max_tokens: int = 4000
+    max_files: int = 10
+    max_lines_per_file: int = 100
+    include_line_numbers: bool = True
+    include_file_content: bool = True
+    format_style: str = "markdown"
+    prioritize_precision: bool = True
+    
+    # Approximate tokens per character (conservative estimate)
+    chars_per_token: float = 4.0
+
+
+class ContextInjector:
+    """Injects FastContext results into agent context.
+    
+    Handles:
+    - Token budget management
+    - Result formatting
+    - Context pollution prevention
+    """
+    
+    def __init__(self, config: Optional[InjectionConfig] = None):
+        """Initialize context injector.
+        
+        Args:
+            config: Injection configuration
+        """
+        self.config = config or InjectionConfig()
+    
+    def estimate_tokens(self, text: str) -> int:
+        """Estimate token count for text.
+        
+        Args:
+            text: Text to estimate
+            
+        Returns:
+            Estimated token count
+        """
+        return int(len(text) / self.config.chars_per_token)
+    
+    def inject(
+        self,
+        result: FastContextResult,
+        system_prompt: Optional[str] = None,
+        user_message: Optional[str] = None
+    ) -> Dict[str, str]:
+        """Inject FastContext results into agent context.
+        
+        Args:
+            result: FastContext search result
+            system_prompt: Original system prompt
+            user_message: Original user message
+            
+        Returns:
+            Dict with 'system_prompt' and 'user_message' keys
+        """
+        # Format the context
+        context_str = self.format_context(result)
+        
+        # Inject into appropriate location
+        if system_prompt:
+            # Add to system prompt
+            injected_system = self._inject_into_system_prompt(system_prompt, context_str)
+            return {
+                "system_prompt": injected_system,
+                "user_message": user_message or ""
+            }
+        elif user_message:
+            # Add to user message
+            injected_user = self._inject_into_user_message(user_message, context_str)
+            return {
+                "system_prompt": "",
+                "user_message": injected_user
+            }
+        else:
+            return {
+                "system_prompt": context_str,
+                "user_message": ""
+            }
+    
+    def format_context(self, result: FastContextResult) -> str:
+        """Format FastContext result as context string.
+        
+        Args:
+            result: FastContext search result
+            
+        Returns:
+            Formatted context string
+        """
+        if self.config.format_style == "markdown":
+            return self._format_markdown(result)
+        elif self.config.format_style == "xml":
+            return self._format_xml(result)
+        else:
+            return self._format_plain(result)
+    
+    def _format_markdown(self, result: FastContextResult) -> str:
+        """Format result as markdown.
+        
+        Args:
+            result: FastContext search result
+            
+        Returns:
+            Markdown formatted string
+        """
+        if not result.files:
+            return ""
+        
+        lines = ["# Relevant Code Context"]
+        lines.append(f"*Found {result.total_files} file(s) in {result.search_time_ms}ms*\n")
+        
+        token_budget = self.config.max_tokens
+        files_included = 0
+        
+        # Sort by relevance
+        sorted_files = sorted(
+            result.files,
+            key=lambda f: f.relevance_score,
+            reverse=True
+        )
+        
+        # Apply precision filter if enabled
+        if self.config.prioritize_precision:
+            sorted_files = [f for f in sorted_files if f.relevance_score >= 0.5]
+        
+        for file_match in sorted_files:
+            if files_included >= self.config.max_files:
+                break
+            
+            file_section = self._format_file_markdown(file_match)
+            section_tokens = self.estimate_tokens(file_section)
+            
+            if section_tokens > token_budget:
+                # Try truncated version
+                file_section = self._format_file_markdown(file_match, truncate=True)
+                section_tokens = self.estimate_tokens(file_section)
+                
+                if section_tokens > token_budget:
+                    continue
+            
+            lines.append(file_section)
+            token_budget -= section_tokens
+            files_included += 1
+        
+        if files_included < result.total_files:
+            lines.append(f"\n*... and {result.total_files - files_included} more files*")
+        
+        return "\n".join(lines)
+    
+    def _format_file_markdown(
+        self,
+        file_match: FileMatch,
+        truncate: bool = False
+    ) -> str:
+        """Format a single file match as markdown.
+        
+        Args:
+            file_match: File match to format
+            truncate: Whether to truncate content
+            
+        Returns:
+            Markdown formatted string
+        """
+        lines = [f"## `{file_match.path}`"]
+        
+        if not self.config.include_file_content:
+            if file_match.line_ranges:
+                ranges = ", ".join(
+                    f"L{lr.start}-{lr.end}" for lr in file_match.line_ranges[:3]
+                )
+                lines.append(f"Relevant lines: {ranges}")
+            return "\n".join(lines)
+        
+        max_lines = self.config.max_lines_per_file
+        if truncate:
+            max_lines = max_lines // 2
+        
+        lines_shown = 0
+        for lr in file_match.line_ranges:
+            if lines_shown >= max_lines:
+                break
+            
+            if self.config.include_line_numbers:
+                lines.append(f"Lines {lr.start}-{lr.end}:")
+            
+            if lr.content:
+                content = lr.content
+                content_lines = content.split("\n")
+                
+                if len(content_lines) > max_lines - lines_shown:
+                    content_lines = content_lines[:max_lines - lines_shown]
+                    content = "\n".join(content_lines) + "\n..."
+                
+                lines.append(f"```\n{content}\n```")
+                lines_shown += len(content_lines)
+        
+        return "\n".join(lines)
+    
+    def _format_xml(self, result: FastContextResult) -> str:
+        """Format result as XML.
+        
+        Args:
+            result: FastContext search result
+            
+        Returns:
+            XML formatted string
+        """
+        if not result.files:
+            return ""
+        
+        lines = ["<code_context>"]
+        lines.append(f"  <summary files=\"{result.total_files}\" search_time_ms=\"{result.search_time_ms}\"/>")
+        
+        token_budget = self.config.max_tokens
+        files_included = 0
+        
+        sorted_files = sorted(
+            result.files,
+            key=lambda f: f.relevance_score,
+            reverse=True
+        )
+        
+        for file_match in sorted_files[:self.config.max_files]:
+            if files_included >= self.config.max_files:
+                break
+            
+            file_xml = self._format_file_xml(file_match)
+            section_tokens = self.estimate_tokens(file_xml)
+            
+            if section_tokens > token_budget:
+                continue
+            
+            lines.append(file_xml)
+            token_budget -= section_tokens
+            files_included += 1
+        
+        lines.append("</code_context>")
+        return "\n".join(lines)
+    
+    def _format_file_xml(self, file_match: FileMatch) -> str:
+        """Format a single file match as XML.
+        
+        Args:
+            file_match: File match to format
+            
+        Returns:
+            XML formatted string
+        """
+        lines = [f'  <file path="{file_match.path}" relevance="{file_match.relevance_score:.2f}">']
+        
+        for lr in file_match.line_ranges[:3]:
+            lines.append(f'    <lines start="{lr.start}" end="{lr.end}">')
+            if lr.content and self.config.include_file_content:
+                # Escape XML special characters
+                content = lr.content.replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
+                lines.append(f"      <![CDATA[{content}]]>")
+            lines.append("    </lines>")
+        
+        lines.append("  </file>")
+        return "\n".join(lines)
+    
+    def _format_plain(self, result: FastContextResult) -> str:
+        """Format result as plain text.
+        
+        Args:
+            result: FastContext search result
+            
+        Returns:
+            Plain text formatted string
+        """
+        if not result.files:
+            return ""
+        
+        lines = [f"Code Context ({result.total_files} files):"]
+        
+        for file_match in result.files[:self.config.max_files]:
+            lines.append(f"\nFile: {file_match.path}")
+            
+            for lr in file_match.line_ranges[:3]:
+                lines.append(f"  Lines {lr.start}-{lr.end}")
+                if lr.content and self.config.include_file_content:
+                    for line in lr.content.split("\n")[:10]:
+                        lines.append(f"    {line}")
+        
+        return "\n".join(lines)
+    
+    def _inject_into_system_prompt(
+        self,
+        system_prompt: str,
+        context: str
+    ) -> str:
+        """Inject context into system prompt.
+        
+        Args:
+            system_prompt: Original system prompt
+            context: Context to inject
+            
+        Returns:
+            Modified system prompt
+        """
+        if not context:
+            return system_prompt
+        
+        # Add context section
+        return f"{system_prompt}\n\n{context}"
+    
+    def _inject_into_user_message(
+        self,
+        user_message: str,
+        context: str
+    ) -> str:
+        """Inject context into user message.
+        
+        Args:
+            user_message: Original user message
+            context: Context to inject
+            
+        Returns:
+            Modified user message
+        """
+        if not context:
+            return user_message
+        
+        # Prepend context
+        return f"{context}\n\n---\n\n{user_message}"
+
+
+def inject_fast_context(
+    result: FastContextResult,
+    system_prompt: Optional[str] = None,
+    user_message: Optional[str] = None,
+    max_tokens: int = 4000,
+    format_style: str = "markdown"
+) -> Dict[str, str]:
+    """Convenience function to inject FastContext results.
+    
+    Args:
+        result: FastContext search result
+        system_prompt: Original system prompt
+        user_message: Original user message
+        max_tokens: Maximum tokens for context
+        format_style: Format style ('markdown', 'xml', 'plain')
+        
+    Returns:
+        Dict with 'system_prompt' and 'user_message' keys
+    """
+    config = InjectionConfig(
+        max_tokens=max_tokens,
+        format_style=format_style
+    )
+    injector = ContextInjector(config)
+    return injector.inject(result, system_prompt, user_message)

src/praisonai-agents/praisonaiagents/context/fast/fast_context.py

@@ -0,0 +1,384 @@
+"""
+FastContext - High-level API for fast parallel code search.
+
+This module provides the main FastContext class that can be used
+standalone or integrated with the main Agent class.
+
+Example usage:
+    # Standalone usage
+    from praisonaiagents.context.fast import FastContext
+    
+    fc = FastContext(workspace_path="/path/to/project")
+    result = fc.search("find authentication handlers")
+    print(result.to_context_string())
+    
+    # With Agent integration
+    from praisonaiagents import Agent
+    
+    agent = Agent(
+        name="CodeAssistant",
+        fast_context=True,
+        fast_context_path="/path/to/project"
+    )
+"""
+
+import os
+import hashlib
+import json
+import logging
+from typing import Optional, List, Dict, Any
+
+from praisonaiagents.context.fast.result import FastContextResult, FileMatch, LineRange
+from praisonaiagents.context.fast.fast_context_agent import FastContextAgent
+
+logger = logging.getLogger(__name__)
+
+
+class FastContext:
+    """High-level API for fast parallel code search.
+    
+    FastContext provides rapid code search capabilities using:
+    - Parallel tool execution (up to 8 concurrent calls)
+    - Limited turns (max 4) for fast response
+    - Optional LLM-powered intelligent search
+    - Result caching for repeated queries
+    
+    Attributes:
+        workspace_path: Root directory for searches
+        model: LLM model for intelligent search
+        max_turns: Maximum search turns
+        max_parallel: Maximum parallel tool calls
+        cache_enabled: Whether to cache results
+    """
+    
+    # Environment variable names for configuration
+    ENV_MODEL = "FAST_CONTEXT_MODEL"
+    ENV_MAX_TURNS = "FAST_CONTEXT_MAX_TURNS"
+    ENV_MAX_PARALLEL = "FAST_CONTEXT_PARALLELISM"
+    ENV_TIMEOUT = "FAST_CONTEXT_TIMEOUT"
+    ENV_CACHE_ENABLED = "FAST_CONTEXT_CACHE"
+    ENV_CACHE_TTL = "FAST_CONTEXT_CACHE_TTL"
+    
+    def __init__(
+        self,
+        workspace_path: Optional[str] = None,
+        model: Optional[str] = None,
+        max_turns: Optional[int] = None,
+        max_parallel: Optional[int] = None,
+        timeout: Optional[float] = None,
+        cache_enabled: Optional[bool] = None,
+        cache_ttl: Optional[int] = None,
+        verbose: bool = False
+    ):
+        """Initialize FastContext.
+        
+        Configuration can be set via parameters or environment variables.
+        Parameters take precedence over environment variables.
+        
+        Environment variables:
+            FAST_CONTEXT_MODEL: LLM model name (default: gpt-4o-mini)
+            FAST_CONTEXT_MAX_TURNS: Maximum search turns (default: 4)
+            FAST_CONTEXT_PARALLELISM: Max parallel calls (default: 8)
+            FAST_CONTEXT_TIMEOUT: Timeout in seconds (default: 30.0)
+            FAST_CONTEXT_CACHE: Enable caching (default: true)
+            FAST_CONTEXT_CACHE_TTL: Cache TTL in seconds (default: 300)
+        
+        Args:
+            workspace_path: Root directory for searches (defaults to cwd)
+            model: LLM model for intelligent search
+            max_turns: Maximum search turns
+            max_parallel: Maximum parallel tool calls per turn
+            timeout: Timeout per tool call in seconds
+            cache_enabled: Whether to cache search results
+            cache_ttl: Cache time-to-live in seconds
+            verbose: If True, print debug information
+        """
+        self.workspace_path = os.path.abspath(workspace_path or os.getcwd())
+        
+        # Apply environment variable overrides with defaults
+        self.model = model or os.environ.get(self.ENV_MODEL, "gpt-4o-mini")
+        self.max_turns = max_turns if max_turns is not None else int(os.environ.get(self.ENV_MAX_TURNS, "4"))
+        self.max_parallel = max_parallel if max_parallel is not None else int(os.environ.get(self.ENV_MAX_PARALLEL, "8"))
+        self.timeout = timeout if timeout is not None else float(os.environ.get(self.ENV_TIMEOUT, "30.0"))
+        
+        # Cache settings
+        cache_env = os.environ.get(self.ENV_CACHE_ENABLED, "true").lower()
+        self.cache_enabled = cache_enabled if cache_enabled is not None else cache_env in ("true", "1", "yes")
+        self.cache_ttl = cache_ttl if cache_ttl is not None else int(os.environ.get(self.ENV_CACHE_TTL, "300"))
+        
+        self.verbose = verbose
+        
+        # Cache storage
+        self._cache: Dict[str, Dict[str, Any]] = {}
+        
+        # Agent instance (lazy loaded)
+        self._agent: Optional[FastContextAgent] = None
+    
+    def _get_agent(self) -> FastContextAgent:
+        """Get or create FastContextAgent instance."""
+        if self._agent is None:
+            self._agent = FastContextAgent(
+                workspace_path=self.workspace_path,
+                max_turns=self.max_turns,
+                max_parallel=self.max_parallel,
+                model=self.model,
+                timeout=self.timeout,
+                verbose=self.verbose
+            )
+        return self._agent
+    
+    def _get_cache_key(self, query: str, **kwargs) -> str:
+        """Generate cache key for a query."""
+        key_data = {
+            "query": query,
+            "workspace": self.workspace_path,
+            **kwargs
+        }
+        key_str = json.dumps(key_data, sort_keys=True)
+        return hashlib.md5(key_str.encode()).hexdigest()
+    
+    def _get_cached(self, cache_key: str) -> Optional[FastContextResult]:
+        """Get cached result if valid."""
+        if not self.cache_enabled:
+            return None
+        
+        import time
+        cached = self._cache.get(cache_key)
+        if cached:
+            if time.time() - cached["timestamp"] < self.cache_ttl:
+                result = cached["result"]
+                result.from_cache = True
+                return result
+            else:
+                # Expired
+                del self._cache[cache_key]
+        return None
+    
+    def _set_cached(self, cache_key: str, result: FastContextResult) -> None:
+        """Cache a result."""
+        if not self.cache_enabled:
+            return
+        
+        import time
+        self._cache[cache_key] = {
+            "result": result,
+            "timestamp": time.time()
+        }
+    
+    def search(
+        self,
+        query: str,
+        include_patterns: Optional[List[str]] = None,
+        exclude_patterns: Optional[List[str]] = None,
+        use_llm: bool = False
+    ) -> FastContextResult:
+        """Search for code relevant to the query.
+        
+        Args:
+            query: Search query (pattern or natural language)
+            include_patterns: Glob patterns to include (e.g., ["*.py"])
+            exclude_patterns: Glob patterns to exclude
+            use_llm: If True, use LLM for intelligent search
+            
+        Returns:
+            FastContextResult with matching files and lines
+        """
+        # Check cache
+        cache_key = self._get_cache_key(
+            query,
+            include_patterns=include_patterns,
+            exclude_patterns=exclude_patterns,
+            use_llm=use_llm
+        )
+        cached = self._get_cached(cache_key)
+        if cached:
+            if self.verbose:
+                logger.info(f"Cache hit for query: {query}")
+            return cached
+        
+        # Perform search
+        agent = self._get_agent()
+        
+        if use_llm:
+            result = agent.search(query, include_patterns, exclude_patterns)
+        else:
+            result = agent.search_simple(query, include_patterns, exclude_patterns)
+        
+        # Cache result
+        self._set_cached(cache_key, result)
+        
+        return result
+    
+    def search_files(
+        self,
+        pattern: str,
+        max_results: int = 50
+    ) -> FastContextResult:
+        """Search for files matching a glob pattern.
+        
+        Args:
+            pattern: Glob pattern (e.g., "**/*.py", "src/*.js")
+            max_results: Maximum results to return
+            
+        Returns:
+            FastContextResult with matching files
+        """
+        from praisonaiagents.context.fast.search_tools import glob_search
+        
+        result = FastContextResult(query=pattern)
+        result.start_timer()
+        
+        matches = glob_search(
+            self.workspace_path,
+            pattern,
+            max_results=max_results
+        )
+        
+        for match in matches:
+            result.add_file(FileMatch(
+                path=match["path"],
+                relevance_score=1.0
+            ))
+        
+        result.stop_timer()
+        result.turns_used = 1
+        result.total_tool_calls = 1
+        
+        return result
+    
+    def read_context(
+        self,
+        filepath: str,
+        start_line: Optional[int] = None,
+        end_line: Optional[int] = None,
+        context_lines: int = 5
+    ) -> Optional[str]:
+        """Read file content with context.
+        
+        Args:
+            filepath: Path to file (relative to workspace)
+            start_line: Starting line (1-indexed)
+            end_line: Ending line (1-indexed)
+            context_lines: Additional context lines
+            
+        Returns:
+            File content string or None if error
+        """
+        from praisonaiagents.context.fast.search_tools import read_file
+        
+        full_path = filepath
+        if not os.path.isabs(filepath):
+            full_path = os.path.join(self.workspace_path, filepath)
+        
+        result = read_file(
+            full_path,
+            start_line=start_line,
+            end_line=end_line,
+            context_lines=context_lines
+        )
+        
+        if result["success"]:
+            return result["content"]
+        return None
+    
+    def get_context_for_agent(
+        self,
+        query: str,
+        max_files: int = 10,
+        max_lines_per_file: int = 100,
+        include_content: bool = True
+    ) -> str:
+        """Get formatted context string for an agent.
+        
+        This method searches for relevant code and formats it
+        as a context string that can be injected into an agent's
+        system prompt or message.
+        
+        Args:
+            query: Search query
+            max_files: Maximum files to include
+            max_lines_per_file: Maximum lines per file
+            include_content: Whether to include file content
+            
+        Returns:
+            Formatted context string
+        """
+        result = self.search(query)
+        
+        if not result.files:
+            return f"No relevant code found for: {query}"
+        
+        lines = [f"# Relevant Code Context for: {query}"]
+        lines.append(f"Found {result.total_files} relevant file(s) in {result.search_time_ms}ms\n")
+        
+        for i, file_match in enumerate(result.files[:max_files]):
+            lines.append(f"## {file_match.path}")
+            
+            if include_content and file_match.line_ranges:
+                for lr in file_match.line_ranges:
+                    # Limit lines
+                    if lr.line_count > max_lines_per_file:
+                        lr = LineRange(
+                            start=lr.start,
+                            end=lr.start + max_lines_per_file - 1,
+                            content=lr.content,
+                            relevance_score=lr.relevance_score
+                        )
+                    
+                    lines.append(f"Lines {lr.start}-{lr.end}:")
+                    
+                    # Read content if not already present
+                    if lr.content:
+                        lines.append(f"```\n{lr.content}\n```")
+                    else:
+                        content = self.read_context(
+                            file_match.path,
+                            start_line=lr.start,
+                            end_line=lr.end
+                        )
+                        if content:
+                            lines.append(f"```\n{content}\n```")
+            
+            lines.append("")
+        
+        if len(result.files) > max_files:
+            lines.append(f"... and {len(result.files) - max_files} more files")
+        
+        return "\n".join(lines)
+    
+    def clear_cache(self) -> None:
+        """Clear the result cache."""
+        self._cache.clear()
+    
+    def close(self) -> None:
+        """Close resources."""
+        if self._agent:
+            self._agent.close()
+            self._agent = None
+        self._cache.clear()
+    
+    def __enter__(self):
+        return self
+    
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self.close()
+
+
+# Convenience function for quick searches
+def fast_search(
+    query: str,
+    workspace_path: Optional[str] = None,
+    **kwargs
+) -> FastContextResult:
+    """Convenience function for quick code search.
+    
+    Args:
+        query: Search query
+        workspace_path: Root directory (defaults to cwd)
+        **kwargs: Additional arguments for FastContext
+        
+    Returns:
+        FastContextResult with matching files
+    """
+    with FastContext(workspace_path=workspace_path, **kwargs) as fc:
+        return fc.search(query)

src/praisonai-agents/praisonaiagents/context/fast/fast_context_agent.py

@@ -0,0 +1,392 @@
+"""
+FastContextAgent - Specialized subagent for rapid code search.
+
+This agent is designed for fast context retrieval using:
+- Parallel tool execution (up to 8 concurrent calls)
+- Limited turns (max 4) for fast response
+- Lightweight model (gpt-4o-mini by default)
+- Restricted tool set (grep, glob, read, list_dir)
+"""
+
+import os
+import json
+import logging
+from typing import List, Dict, Any, Optional
+
+from praisonaiagents.context.fast.result import (
+    FastContextResult,
+    FileMatch,
+    LineRange
+)
+from praisonaiagents.context.fast.search_tools import (
+    grep_search,
+    glob_search,
+    read_file,
+    list_directory,
+    FAST_CONTEXT_TOOLS
+)
+from praisonaiagents.context.fast.parallel_executor import (
+    ToolCallBatch,
+    ParallelSearchCoordinator
+)
+
+logger = logging.getLogger(__name__)
+
+
+class FastContextAgent:
+    """Specialized agent for fast parallel code search.
+    
+    This agent performs rapid code search using parallel tool calls
+    and a lightweight model optimized for retrieval tasks.
+    
+    Attributes:
+        workspace_path: Root directory for searches
+        max_turns: Maximum search turns (default: 4)
+        max_parallel: Maximum parallel tool calls (default: 8)
+        model: LLM model for intelligent search (default: gpt-4o-mini)
+        timeout: Timeout per tool call in seconds
+    """
+    
+    def __init__(
+        self,
+        workspace_path: str,
+        max_turns: int = 4,
+        max_parallel: int = 8,
+        model: str = "gpt-4o-mini",
+        timeout: float = 30.0,
+        verbose: bool = False
+    ):
+        """Initialize FastContextAgent.
+        
+        Args:
+            workspace_path: Root directory for searches
+            max_turns: Maximum search turns
+            max_parallel: Maximum parallel tool calls per turn
+            model: LLM model name for intelligent search
+            timeout: Timeout per tool call in seconds
+            verbose: If True, print debug information
+        """
+        self.workspace_path = os.path.abspath(workspace_path)
+        self.max_turns = max_turns
+        self.max_parallel = max_parallel
+        self.model = model
+        self.timeout = timeout
+        self.verbose = verbose
+        
+        # Initialize coordinator
+        self.coordinator = ParallelSearchCoordinator(
+            max_parallel=max_parallel,
+            max_turns=max_turns,
+            timeout=timeout
+        )
+        
+        # LLM instance (lazy loaded)
+        self._llm = None
+    
+    def _get_llm(self):
+        """Get or create LLM instance (lazy loading)."""
+        if self._llm is None:
+            try:
+                from praisonaiagents.llm import LLM
+                self._llm = LLM(model=self.model)
+            except Exception as e:
+                logger.warning(f"Could not initialize LLM: {e}")
+                self._llm = None
+        return self._llm
+    
+    def get_tools(self) -> List[Dict[str, Any]]:
+        """Get tool definitions for LLM function calling.
+        
+        Returns:
+            List of tool definition dictionaries
+        """
+        return FAST_CONTEXT_TOOLS
+    
+    def execute_tool(self, tool_name: str, **kwargs) -> Any:
+        """Execute a single tool.
+        
+        Args:
+            tool_name: Name of the tool
+            **kwargs: Tool arguments
+            
+        Returns:
+            Tool result
+        """
+        # Prepend workspace path if needed
+        if tool_name in ("grep_search", "glob_search"):
+            if "search_path" not in kwargs or kwargs["search_path"] == ".":
+                kwargs["search_path"] = self.workspace_path
+            elif not os.path.isabs(kwargs["search_path"]):
+                kwargs["search_path"] = os.path.join(
+                    self.workspace_path, kwargs["search_path"]
+                )
+        elif tool_name == "list_directory":
+            if "dir_path" not in kwargs or kwargs["dir_path"] == ".":
+                kwargs["dir_path"] = self.workspace_path
+            elif not os.path.isabs(kwargs["dir_path"]):
+                kwargs["dir_path"] = os.path.join(
+                    self.workspace_path, kwargs["dir_path"]
+                )
+        elif tool_name == "read_file":
+            if "filepath" in kwargs and not os.path.isabs(kwargs["filepath"]):
+                kwargs["filepath"] = os.path.join(
+                    self.workspace_path, kwargs["filepath"]
+                )
+        
+        tools = {
+            "grep_search": grep_search,
+            "glob_search": glob_search,
+            "read_file": read_file,
+            "list_directory": list_directory
+        }
+        
+        if tool_name not in tools:
+            return {"error": f"Unknown tool: {tool_name}"}
+        
+        try:
+            return tools[tool_name](**kwargs)
+        except Exception as e:
+            return {"error": str(e)}
+    
+    def search_simple(
+        self,
+        query: str,
+        include_patterns: Optional[List[str]] = None,
+        exclude_patterns: Optional[List[str]] = None,
+        max_results: int = 50
+    ) -> FastContextResult:
+        """Perform a simple search without LLM.
+        
+        This is a fast, direct search that doesn't use the LLM
+        for query understanding. Good for simple pattern searches.
+        
+        Args:
+            query: Search pattern
+            include_patterns: Glob patterns to include
+            exclude_patterns: Glob patterns to exclude
+            max_results: Maximum results per tool
+            
+        Returns:
+            FastContextResult with matching files and lines
+        """
+        result = FastContextResult(query=query)
+        result.start_timer()
+        
+        # Execute grep search
+        grep_results = grep_search(
+            search_path=self.workspace_path,
+            pattern=query,
+            include_patterns=include_patterns,
+            exclude_patterns=exclude_patterns,
+            max_results=max_results,
+            context_lines=2
+        )
+        
+        # Convert to FastContextResult format
+        for match in grep_results:
+            file_match = FileMatch(
+                path=match["path"],
+                relevance_score=1.0,
+                match_count=1
+            )
+            file_match.add_line_range(LineRange(
+                start=max(1, match["line_number"] - 2),
+                end=match["line_number"] + 2,
+                content=match.get("context"),
+                relevance_score=1.0
+            ))
+            result.add_file(file_match)
+        
+        result.stop_timer()
+        result.turns_used = 1
+        result.total_tool_calls = 1
+        
+        return result
+    
+    def search(
+        self,
+        query: str,
+        include_patterns: Optional[List[str]] = None,
+        exclude_patterns: Optional[List[str]] = None
+    ) -> FastContextResult:
+        """Perform intelligent search using LLM.
+        
+        Uses the LLM to understand the query and plan search strategy,
+        then executes parallel tool calls to find relevant code.
+        
+        Args:
+            query: Natural language search query
+            include_patterns: Glob patterns to include
+            exclude_patterns: Glob patterns to exclude
+            
+        Returns:
+            FastContextResult with matching files and lines
+        """
+        result = FastContextResult(query=query)
+        result.start_timer()
+        
+        llm = self._get_llm()
+        
+        # If no LLM available, fall back to simple search
+        if llm is None:
+            logger.info("LLM not available, falling back to simple search")
+            return self.search_simple(query, include_patterns, exclude_patterns)
+        
+        self.coordinator.start()
+        
+        # System prompt for the search agent
+        system_prompt = f"""You are a fast code search agent. Your task is to find relevant code in a codebase.
+
+Workspace: {self.workspace_path}
+
+You have access to these tools:
+- grep_search: Search for patterns in files
+- glob_search: Find files by pattern
+- read_file: Read file contents
+- list_directory: List directory contents
+
+Strategy:
+1. Start with broad searches (glob, list_directory) to understand structure
+2. Use grep_search to find specific patterns
+3. Use read_file to get detailed context for relevant files
+4. Execute up to {self.max_parallel} tool calls in parallel per turn
+5. Complete search in {self.max_turns} turns or less
+
+Return files and line ranges that are relevant to the query.
+Be efficient - use parallel tool calls to explore multiple paths at once."""
+
+        messages = [
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": f"Find code relevant to: {query}"}
+        ]
+        
+        # Add file pattern constraints if provided
+        if include_patterns:
+            messages[1]["content"] += f"\nOnly search in files matching: {include_patterns}"
+        if exclude_patterns:
+            messages[1]["content"] += f"\nExclude files matching: {exclude_patterns}"
+        
+        try:
+            # Multi-turn search loop
+            for turn in range(self.max_turns):
+                if self.verbose:
+                    logger.info(f"Search turn {turn + 1}/{self.max_turns}")
+                
+                # Get LLM response with tool calls
+                response = llm.client.chat.completions.create(
+                    model=self.model,
+                    messages=messages,
+                    tools=[{"type": "function", "function": t} for t in self.get_tools()],
+                    tool_choice="auto"
+                )
+                
+                message = response.choices[0].message
+                
+                # Check if we have tool calls
+                if not message.tool_calls:
+                    # No more tool calls, search complete
+                    break
+                
+                # Execute tool calls in parallel
+                batch = ToolCallBatch(max_size=self.max_parallel)
+                tool_call_map = {}  # Map tool call ID to batch index
+                
+                for tool_call in message.tool_calls[:self.max_parallel]:
+                    try:
+                        args = json.loads(tool_call.function.arguments)
+                        batch.add(tool_call.function.name, **args)
+                        tool_call_map[tool_call.id] = len(batch.tasks) - 1
+                    except json.JSONDecodeError:
+                        continue
+                
+                # Execute batch
+                if batch.tasks:
+                    tool_results = self.coordinator.execute_turn_sync(batch)
+                    
+                    # Process results
+                    messages.append({"role": "assistant", "content": None, "tool_calls": [
+                        {
+                            "id": tc.id,
+                            "type": "function",
+                            "function": {
+                                "name": tc.function.name,
+                                "arguments": tc.function.arguments
+                            }
+                        }
+                        for tc in message.tool_calls[:self.max_parallel]
+                    ]})
+                    
+                    # Add tool results to messages
+                    for tool_call in message.tool_calls[:self.max_parallel]:
+                        idx = tool_call_map.get(tool_call.id)
+                        if idx is not None and idx < len(tool_results):
+                            tool_result = tool_results[idx]
+                            
+                            # Process grep results into FastContextResult
+                            if tool_call.function.name == "grep_search":
+                                if isinstance(tool_result, list):
+                                    for match in tool_result:
+                                        file_match = FileMatch(
+                                            path=match.get("path", ""),
+                                            relevance_score=1.0,
+                                            match_count=1
+                                        )
+                                        line_num = match.get("line_number", 1)
+                                        file_match.add_line_range(LineRange(
+                                            start=max(1, line_num - 2),
+                                            end=line_num + 2,
+                                            content=match.get("context"),
+                                            relevance_score=1.0
+                                        ))
+                                        result.add_file(file_match)
+                            
+                            messages.append({
+                                "role": "tool",
+                                "tool_call_id": tool_call.id,
+                                "content": json.dumps(tool_result) if not isinstance(tool_result, str) else tool_result
+                            })
+        
+        except Exception as e:
+            logger.error(f"Error during LLM search: {e}")
+            # Fall back to simple search
+            return self.search_simple(query, include_patterns, exclude_patterns)
+        
+        finally:
+            self.coordinator.stop()
+        
+        result.stop_timer()
+        result.turns_used = self.coordinator.turns_used
+        result.total_tool_calls = self.coordinator.total_tool_calls
+        result.sort_by_relevance()
+        
+        return result
+    
+    async def search_async(
+        self,
+        query: str,
+        include_patterns: Optional[List[str]] = None,
+        exclude_patterns: Optional[List[str]] = None
+    ) -> FastContextResult:
+        """Async version of search.
+        
+        Args:
+            query: Natural language search query
+            include_patterns: Glob patterns to include
+            exclude_patterns: Glob patterns to exclude
+            
+        Returns:
+            FastContextResult with matching files and lines
+        """
+        # For now, wrap sync version
+        # TODO: Implement true async with async LLM calls
+        return self.search(query, include_patterns, exclude_patterns)
+    
+    def close(self) -> None:
+        """Close resources."""
+        self.coordinator.close()
+    
+    def __enter__(self):
+        return self
+    
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self.close()

src/praisonai-agents/praisonaiagents/context/fast/indexer/__init__.py

@@ -0,0 +1,23 @@
+"""
+Indexer module for Fast Context.
+
+Provides file and symbol indexing capabilities for fast code search.
+"""
+
+def __getattr__(name):
+    """Lazy load indexer classes."""
+    if name == "FileIndexer":
+        from .file_indexer import FileIndexer
+        return FileIndexer
+    elif name == "SymbolIndexer":
+        from .symbol_indexer import SymbolIndexer
+        return SymbolIndexer
+    elif name == "Symbol":
+        from .symbol_indexer import Symbol
+        return Symbol
+    elif name == "SymbolType":
+        from .symbol_indexer import SymbolType
+        return SymbolType
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+__all__ = ["FileIndexer", "SymbolIndexer", "Symbol", "SymbolType"]

src/praisonai-agents/praisonaiagents/context/fast/indexer/file_indexer.py

@@ -0,0 +1,445 @@
+"""
+File Indexer for Fast Context.
+
+Provides file indexing capabilities for faster code search:
+- Indexes file paths and metadata
+- Supports .gitignore and .praisonignore patterns
+- Incremental updates on file changes
+- Memory-efficient storage
+"""
+
+import os
+import fnmatch
+import hashlib
+import json
+import logging
+from dataclasses import dataclass
+from typing import Dict, List, Set, Optional, Any
+from pathlib import Path
+import time
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class FileInfo:
+    """Information about an indexed file.
+    
+    Attributes:
+        path: Relative path from workspace root
+        absolute_path: Absolute file path
+        size: File size in bytes
+        modified: Last modification timestamp
+        extension: File extension (lowercase)
+        hash: Content hash for change detection
+    """
+    path: str
+    absolute_path: str
+    size: int
+    modified: float
+    extension: str
+    hash: Optional[str] = None
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary."""
+        return {
+            "path": self.path,
+            "absolute_path": self.absolute_path,
+            "size": self.size,
+            "modified": self.modified,
+            "extension": self.extension,
+            "hash": self.hash
+        }
+    
+    @classmethod
+    def from_dict(cls, data: Dict[str, Any]) -> "FileInfo":
+        """Create from dictionary."""
+        return cls(**data)
+
+
+class FileIndexer:
+    """Indexes files in a workspace for fast search.
+    
+    The indexer maintains an in-memory index of files that can be
+    used to quickly find files by pattern without filesystem traversal.
+    
+    Attributes:
+        workspace_path: Root directory to index
+        files: Dictionary mapping relative paths to FileInfo
+        extensions: Set of indexed file extensions
+    """
+    
+    # Default extensions to index
+    DEFAULT_EXTENSIONS = {
+        ".py", ".js", ".ts", ".jsx", ".tsx", ".java", ".go", ".rs",
+        ".c", ".cpp", ".h", ".hpp", ".cs", ".rb", ".php", ".swift",
+        ".kt", ".scala", ".r", ".m", ".mm", ".sql", ".sh", ".bash",
+        ".yaml", ".yml", ".json", ".xml", ".html", ".css", ".scss",
+        ".md", ".txt", ".rst", ".toml", ".ini", ".cfg", ".conf"
+    }
+    
+    # Default patterns to ignore
+    DEFAULT_IGNORE_PATTERNS = {
+        "__pycache__", "*.pyc", "*.pyo", ".git", ".svn", ".hg",
+        "node_modules", "venv", ".venv", "env", ".env",
+        "dist", "build", ".cache", ".pytest_cache",
+        "*.egg-info", "*.egg", "*.whl", "*.so", "*.dll", "*.dylib"
+    }
+    
+    def __init__(
+        self,
+        workspace_path: str,
+        extensions: Optional[Set[str]] = None,
+        ignore_patterns: Optional[Set[str]] = None,
+        respect_gitignore: bool = True,
+        max_file_size: int = 10 * 1024 * 1024,  # 10MB
+        compute_hashes: bool = False
+    ):
+        """Initialize file indexer.
+        
+        Args:
+            workspace_path: Root directory to index
+            extensions: File extensions to index (None = all text files)
+            ignore_patterns: Patterns to ignore
+            respect_gitignore: Whether to respect .gitignore
+            max_file_size: Maximum file size to index
+            compute_hashes: Whether to compute content hashes
+        """
+        self.workspace_path = os.path.abspath(workspace_path)
+        self.extensions = extensions or self.DEFAULT_EXTENSIONS
+        self.ignore_patterns = ignore_patterns or self.DEFAULT_IGNORE_PATTERNS.copy()
+        self.respect_gitignore = respect_gitignore
+        self.max_file_size = max_file_size
+        self.compute_hashes = compute_hashes
+        
+        # Index storage
+        self.files: Dict[str, FileInfo] = {}
+        self.by_extension: Dict[str, List[str]] = {}
+        
+        # Gitignore patterns
+        self._gitignore_patterns: Set[str] = set()
+        
+        # Index metadata
+        self.last_indexed: Optional[float] = None
+        self.total_files: int = 0
+        self.total_size: int = 0
+    
+    def _load_gitignore(self) -> None:
+        """Load patterns from .gitignore and .praisonignore."""
+        self._gitignore_patterns.clear()
+        
+        for filename in [".gitignore", ".praisonignore"]:
+            filepath = os.path.join(self.workspace_path, filename)
+            if os.path.exists(filepath):
+                try:
+                    with open(filepath, 'r', encoding='utf-8', errors='ignore') as f:
+                        for line in f:
+                            line = line.strip()
+                            if line and not line.startswith('#'):
+                                self._gitignore_patterns.add(line)
+                except Exception:
+                    pass
+    
+    def _should_ignore(self, path: str, is_dir: bool = False) -> bool:
+        """Check if a path should be ignored.
+        
+        Args:
+            path: Relative path from workspace
+            is_dir: Whether the path is a directory
+            
+        Returns:
+            True if path should be ignored
+        """
+        name = os.path.basename(path)
+        parts = Path(path).parts
+        
+        # Check default ignore patterns
+        for pattern in self.ignore_patterns:
+            if fnmatch.fnmatch(name, pattern):
+                return True
+            if any(fnmatch.fnmatch(part, pattern) for part in parts):
+                return True
+        
+        # Check gitignore patterns
+        if self.respect_gitignore:
+            for pattern in self._gitignore_patterns:
+                # Handle directory patterns
+                if pattern.endswith('/'):
+                    dir_pattern = pattern.rstrip('/')
+                    if is_dir and fnmatch.fnmatch(name, dir_pattern):
+                        return True
+                    if any(fnmatch.fnmatch(part, dir_pattern) for part in parts):
+                        return True
+                # Handle ** patterns
+                elif '**' in pattern:
+                    fnmatch_pattern = pattern.replace('**/', '*').replace('**', '*')
+                    if fnmatch.fnmatch(path, fnmatch_pattern):
+                        return True
+                # Handle simple patterns
+                else:
+                    if fnmatch.fnmatch(name, pattern):
+                        return True
+                    if fnmatch.fnmatch(path, pattern):
+                        return True
+        
+        return False
+    
+    def _compute_hash(self, filepath: str) -> Optional[str]:
+        """Compute MD5 hash of file content.
+        
+        Args:
+            filepath: Path to file
+            
+        Returns:
+            MD5 hash string or None on error
+        """
+        if not self.compute_hashes:
+            return None
+        
+        try:
+            hasher = hashlib.md5()
+            with open(filepath, 'rb') as f:
+                for chunk in iter(lambda: f.read(8192), b''):
+                    hasher.update(chunk)
+            return hasher.hexdigest()
+        except Exception:
+            return None
+    
+    def index(self) -> int:
+        """Index all files in the workspace.
+        
+        Returns:
+            Number of files indexed
+        """
+        start_time = time.perf_counter()
+        
+        # Load gitignore patterns
+        if self.respect_gitignore:
+            self._load_gitignore()
+        
+        # Clear existing index
+        self.files.clear()
+        self.by_extension.clear()
+        self.total_files = 0
+        self.total_size = 0
+        
+        # Walk directory tree
+        for root, dirs, files in os.walk(self.workspace_path):
+            # Get relative path
+            rel_root = os.path.relpath(root, self.workspace_path)
+            if rel_root == '.':
+                rel_root = ''
+            
+            # Filter directories in-place
+            dirs[:] = [d for d in dirs if not self._should_ignore(
+                os.path.join(rel_root, d) if rel_root else d, is_dir=True
+            )]
+            
+            for filename in files:
+                rel_path = os.path.join(rel_root, filename) if rel_root else filename
+                abs_path = os.path.join(root, filename)
+                
+                # Check if should ignore
+                if self._should_ignore(rel_path):
+                    continue
+                
+                # Check extension
+                ext = os.path.splitext(filename)[1].lower()
+                if self.extensions and ext not in self.extensions:
+                    continue
+                
+                try:
+                    stat = os.stat(abs_path)
+                    
+                    # Check file size
+                    if stat.st_size > self.max_file_size:
+                        continue
+                    
+                    # Create file info
+                    file_info = FileInfo(
+                        path=rel_path,
+                        absolute_path=abs_path,
+                        size=stat.st_size,
+                        modified=stat.st_mtime,
+                        extension=ext,
+                        hash=self._compute_hash(abs_path)
+                    )
+                    
+                    # Add to index
+                    self.files[rel_path] = file_info
+                    
+                    # Add to extension index
+                    if ext not in self.by_extension:
+                        self.by_extension[ext] = []
+                    self.by_extension[ext].append(rel_path)
+                    
+                    self.total_files += 1
+                    self.total_size += stat.st_size
+                    
+                except Exception as e:
+                    logger.debug(f"Error indexing {abs_path}: {e}")
+                    continue
+        
+        self.last_indexed = time.perf_counter()
+        elapsed = self.last_indexed - start_time
+        
+        logger.info(f"Indexed {self.total_files} files ({self.total_size / 1024 / 1024:.2f} MB) in {elapsed:.2f}s")
+        
+        return self.total_files
+    
+    def find_by_pattern(self, pattern: str) -> List[FileInfo]:
+        """Find files matching a glob pattern.
+        
+        Args:
+            pattern: Glob pattern (e.g., "**/*.py", "src/*.js")
+            
+        Returns:
+            List of matching FileInfo objects
+        """
+        results = []
+        
+        for rel_path, file_info in self.files.items():
+            if fnmatch.fnmatch(rel_path, pattern):
+                results.append(file_info)
+            elif fnmatch.fnmatch(os.path.basename(rel_path), pattern):
+                results.append(file_info)
+        
+        return results
+    
+    def find_by_extension(self, extension: str) -> List[FileInfo]:
+        """Find files by extension.
+        
+        Args:
+            extension: File extension (with or without dot)
+            
+        Returns:
+            List of matching FileInfo objects
+        """
+        if not extension.startswith('.'):
+            extension = '.' + extension
+        extension = extension.lower()
+        
+        paths = self.by_extension.get(extension, [])
+        return [self.files[p] for p in paths if p in self.files]
+    
+    def find_by_name(self, name: str, exact: bool = False) -> List[FileInfo]:
+        """Find files by name.
+        
+        Args:
+            name: Filename to search for
+            exact: If True, match exact name; otherwise, partial match
+            
+        Returns:
+            List of matching FileInfo objects
+        """
+        results = []
+        name_lower = name.lower()
+        
+        for rel_path, file_info in self.files.items():
+            filename = os.path.basename(rel_path).lower()
+            if exact:
+                if filename == name_lower:
+                    results.append(file_info)
+            else:
+                if name_lower in filename:
+                    results.append(file_info)
+        
+        return results
+    
+    def get_file(self, path: str) -> Optional[FileInfo]:
+        """Get file info by path.
+        
+        Args:
+            path: Relative path from workspace
+            
+        Returns:
+            FileInfo or None if not found
+        """
+        return self.files.get(path)
+    
+    def is_stale(self, max_age: float = 60.0) -> bool:
+        """Check if index is stale.
+        
+        Args:
+            max_age: Maximum age in seconds
+            
+        Returns:
+            True if index is stale or not indexed
+        """
+        if self.last_indexed is None:
+            return True
+        return (time.perf_counter() - self.last_indexed) > max_age
+    
+    def get_stats(self) -> Dict[str, Any]:
+        """Get index statistics.
+        
+        Returns:
+            Dictionary with index stats
+        """
+        return {
+            "workspace_path": self.workspace_path,
+            "total_files": self.total_files,
+            "total_size_bytes": self.total_size,
+            "total_size_mb": round(self.total_size / 1024 / 1024, 2),
+            "extensions": dict(sorted(
+                ((ext, len(paths)) for ext, paths in self.by_extension.items()),
+                key=lambda x: x[1],
+                reverse=True
+            )),
+            "last_indexed": self.last_indexed
+        }
+    
+    def save_index(self, filepath: str) -> None:
+        """Save index to file.
+        
+        Args:
+            filepath: Path to save index
+        """
+        data = {
+            "workspace_path": self.workspace_path,
+            "last_indexed": self.last_indexed,
+            "files": {path: info.to_dict() for path, info in self.files.items()}
+        }
+        
+        with open(filepath, 'w', encoding='utf-8') as f:
+            json.dump(data, f, indent=2)
+    
+    def load_index(self, filepath: str) -> bool:
+        """Load index from file.
+        
+        Args:
+            filepath: Path to load index from
+            
+        Returns:
+            True if loaded successfully
+        """
+        try:
+            with open(filepath, 'r', encoding='utf-8') as f:
+                data = json.load(f)
+            
+            # Verify workspace path matches
+            if data.get("workspace_path") != self.workspace_path:
+                return False
+            
+            self.last_indexed = data.get("last_indexed")
+            self.files.clear()
+            self.by_extension.clear()
+            self.total_files = 0
+            self.total_size = 0
+            
+            for path, info_dict in data.get("files", {}).items():
+                file_info = FileInfo.from_dict(info_dict)
+                self.files[path] = file_info
+                
+                ext = file_info.extension
+                if ext not in self.by_extension:
+                    self.by_extension[ext] = []
+                self.by_extension[ext].append(path)
+                
+                self.total_files += 1
+                self.total_size += file_info.size
+            
+            return True
+        except Exception as e:
+            logger.debug(f"Error loading index: {e}")
+            return False

src/praisonai-agents/praisonaiagents/context/fast/indexer/file_watcher.py

@@ -0,0 +1,321 @@
+"""
+File Watcher for Incremental Indexing.
+
+Monitors file system changes and triggers incremental index updates:
+- Watches for file creation, modification, deletion
+- Debounces rapid changes
+- Supports background indexing
+"""
+
+import os
+import time
+import logging
+import threading
+from typing import Dict, Set, Callable, Optional, Any
+from dataclasses import dataclass
+from enum import Enum
+
+logger = logging.getLogger(__name__)
+
+
+class ChangeType(Enum):
+    """Types of file system changes."""
+    CREATED = "created"
+    MODIFIED = "modified"
+    DELETED = "deleted"
+
+
+@dataclass
+class FileChange:
+    """Represents a file system change.
+    
+    Attributes:
+        path: Path to the changed file
+        change_type: Type of change
+        timestamp: When the change was detected
+    """
+    path: str
+    change_type: ChangeType
+    timestamp: float
+
+
+class FileWatcher:
+    """Watches for file system changes and triggers callbacks.
+    
+    Uses polling-based approach for cross-platform compatibility.
+    For production use, consider using watchdog library.
+    
+    Attributes:
+        workspace_path: Root directory to watch
+        extensions: File extensions to watch
+        poll_interval: Seconds between polls
+    """
+    
+    def __init__(
+        self,
+        workspace_path: str,
+        extensions: Optional[Set[str]] = None,
+        poll_interval: float = 1.0,
+        debounce_delay: float = 0.5,
+        on_change: Optional[Callable[[FileChange], None]] = None
+    ):
+        """Initialize file watcher.
+        
+        Args:
+            workspace_path: Root directory to watch
+            extensions: File extensions to watch (None = all)
+            poll_interval: Seconds between polls
+            debounce_delay: Delay before processing changes
+            on_change: Callback for file changes
+        """
+        self.workspace_path = os.path.abspath(workspace_path)
+        self.extensions = extensions or {
+            '.py', '.js', '.ts', '.jsx', '.tsx', '.go', '.rs', '.java',
+            '.c', '.cpp', '.h', '.hpp', '.cs', '.rb', '.php', '.swift'
+        }
+        self.poll_interval = poll_interval
+        self.debounce_delay = debounce_delay
+        self.on_change = on_change
+        
+        # State tracking
+        self._file_mtimes: Dict[str, float] = {}
+        self._pending_changes: Dict[str, FileChange] = {}
+        self._running = False
+        self._thread: Optional[threading.Thread] = None
+        self._lock = threading.Lock()
+    
+    def _scan_files(self) -> Dict[str, float]:
+        """Scan workspace and get file modification times.
+        
+        Returns:
+            Dict mapping file paths to modification times
+        """
+        mtimes = {}
+        
+        for root, dirs, files in os.walk(self.workspace_path):
+            # Skip hidden directories
+            dirs[:] = [d for d in dirs if not d.startswith('.')]
+            
+            for filename in files:
+                ext = os.path.splitext(filename)[1].lower()
+                if ext in self.extensions:
+                    file_path = os.path.join(root, filename)
+                    try:
+                        mtimes[file_path] = os.path.getmtime(file_path)
+                    except OSError:
+                        pass
+        
+        return mtimes
+    
+    def _detect_changes(self, new_mtimes: Dict[str, float]) -> None:
+        """Detect changes between old and new file states.
+        
+        Args:
+            new_mtimes: New file modification times
+        """
+        old_paths = set(self._file_mtimes.keys())
+        new_paths = set(new_mtimes.keys())
+        
+        now = time.time()
+        
+        # Detect created files
+        for path in new_paths - old_paths:
+            self._add_pending_change(FileChange(
+                path=path,
+                change_type=ChangeType.CREATED,
+                timestamp=now
+            ))
+        
+        # Detect deleted files
+        for path in old_paths - new_paths:
+            self._add_pending_change(FileChange(
+                path=path,
+                change_type=ChangeType.DELETED,
+                timestamp=now
+            ))
+        
+        # Detect modified files
+        for path in old_paths & new_paths:
+            if new_mtimes[path] > self._file_mtimes[path]:
+                self._add_pending_change(FileChange(
+                    path=path,
+                    change_type=ChangeType.MODIFIED,
+                    timestamp=now
+                ))
+        
+        self._file_mtimes = new_mtimes
+    
+    def _add_pending_change(self, change: FileChange) -> None:
+        """Add a change to pending queue (with debouncing).
+        
+        Args:
+            change: File change to add
+        """
+        with self._lock:
+            self._pending_changes[change.path] = change
+    
+    def _process_pending_changes(self) -> None:
+        """Process pending changes after debounce delay."""
+        with self._lock:
+            now = time.time()
+            to_process = []
+            
+            for path, change in list(self._pending_changes.items()):
+                if now - change.timestamp >= self.debounce_delay:
+                    to_process.append(change)
+                    del self._pending_changes[path]
+        
+        # Process changes outside lock
+        for change in to_process:
+            if self.on_change:
+                try:
+                    self.on_change(change)
+                except Exception as e:
+                    logger.error(f"Error processing change {change.path}: {e}")
+    
+    def _poll_loop(self) -> None:
+        """Main polling loop."""
+        # Initial scan
+        self._file_mtimes = self._scan_files()
+        
+        while self._running:
+            try:
+                # Scan for changes
+                new_mtimes = self._scan_files()
+                self._detect_changes(new_mtimes)
+                
+                # Process pending changes
+                self._process_pending_changes()
+                
+            except Exception as e:
+                logger.error(f"Error in file watcher: {e}")
+            
+            time.sleep(self.poll_interval)
+    
+    def start(self) -> None:
+        """Start watching for file changes."""
+        if self._running:
+            return
+        
+        self._running = True
+        self._thread = threading.Thread(target=self._poll_loop, daemon=True)
+        self._thread.start()
+        logger.info(f"File watcher started for {self.workspace_path}")
+    
+    def stop(self) -> None:
+        """Stop watching for file changes."""
+        self._running = False
+        if self._thread:
+            self._thread.join(timeout=2.0)
+            self._thread = None
+        logger.info("File watcher stopped")
+    
+    def __enter__(self):
+        self.start()
+        return self
+    
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self.stop()
+
+
+class IncrementalIndexer:
+    """Manages incremental index updates based on file changes.
+    
+    Combines FileIndexer/SymbolIndexer with FileWatcher for
+    automatic index updates when files change.
+    """
+    
+    def __init__(
+        self,
+        workspace_path: str,
+        file_indexer: Optional[Any] = None,
+        symbol_indexer: Optional[Any] = None,
+        poll_interval: float = 2.0
+    ):
+        """Initialize incremental indexer.
+        
+        Args:
+            workspace_path: Root directory to index
+            file_indexer: FileIndexer instance (optional)
+            symbol_indexer: SymbolIndexer instance (optional)
+            poll_interval: Seconds between file system polls
+        """
+        self.workspace_path = os.path.abspath(workspace_path)
+        self.file_indexer = file_indexer
+        self.symbol_indexer = symbol_indexer
+        
+        self._watcher = FileWatcher(
+            workspace_path=workspace_path,
+            poll_interval=poll_interval,
+            on_change=self._handle_change
+        )
+        
+        # Statistics
+        self.changes_processed = 0
+        self.last_change_time: Optional[float] = None
+    
+    def _handle_change(self, change: FileChange) -> None:
+        """Handle a file change.
+        
+        Args:
+            change: File change to process
+        """
+        rel_path = os.path.relpath(change.path, self.workspace_path)
+        logger.debug(f"Processing {change.change_type.value}: {rel_path}")
+        
+        if change.change_type == ChangeType.DELETED:
+            # Remove from indexes
+            if self.file_indexer and hasattr(self.file_indexer, 'files'):
+                self.file_indexer.files.pop(rel_path, None)
+            if self.symbol_indexer and hasattr(self.symbol_indexer, 'symbols'):
+                self.symbol_indexer.symbols.pop(rel_path, None)
+        else:
+            # Re-index the file
+            if self.file_indexer and hasattr(self.file_indexer, 'index_file'):
+                # FileIndexer doesn't have index_file, so we skip
+                pass
+            if self.symbol_indexer and hasattr(self.symbol_indexer, 'index_file'):
+                self.symbol_indexer.index_file(change.path)
+        
+        self.changes_processed += 1
+        self.last_change_time = time.time()
+    
+    def start(self) -> None:
+        """Start incremental indexing."""
+        # Initial full index
+        if self.file_indexer:
+            self.file_indexer.index()
+        if self.symbol_indexer:
+            self.symbol_indexer.index()
+        
+        # Start watching
+        self._watcher.start()
+    
+    def stop(self) -> None:
+        """Stop incremental indexing."""
+        self._watcher.stop()
+    
+    def get_stats(self) -> Dict[str, Any]:
+        """Get indexer statistics.
+        
+        Returns:
+            Dictionary with stats
+        """
+        stats = {
+            "changes_processed": self.changes_processed,
+            "last_change_time": self.last_change_time
+        }
+        
+        if self.file_indexer:
+            stats["file_indexer"] = self.file_indexer.get_stats()
+        if self.symbol_indexer:
+            stats["symbol_indexer"] = self.symbol_indexer.get_stats()
+        
+        return stats
+    
+    def __enter__(self):
+        self.start()
+        return self
+    
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self.stop()

src/praisonai-agents/praisonaiagents/context/fast/indexer/symbol_indexer.py

@@ -0,0 +1,669 @@
+"""
+Symbol Indexer for Fast Context.
+
+Extracts code symbols from source files:
+- Python: functions, classes, imports
+- JavaScript/TypeScript: functions, classes, imports, exports
+- Go: functions, types, imports
+- Rust: functions, structs, impl blocks, use statements
+- Java: classes, methods, imports
+"""
+
+import os
+import re
+import logging
+from dataclasses import dataclass, field
+from typing import Dict, List, Set, Optional, Any
+from pathlib import Path
+from enum import Enum
+
+logger = logging.getLogger(__name__)
+
+
+class SymbolType(Enum):
+    """Types of code symbols."""
+    FUNCTION = "function"
+    CLASS = "class"
+    METHOD = "method"
+    IMPORT = "import"
+    EXPORT = "export"
+    VARIABLE = "variable"
+    CONSTANT = "constant"
+    TYPE = "type"
+    STRUCT = "struct"
+    INTERFACE = "interface"
+    MODULE = "module"
+
+
+@dataclass
+class Symbol:
+    """Represents a code symbol.
+    
+    Attributes:
+        name: Symbol name
+        symbol_type: Type of symbol
+        file_path: File containing the symbol
+        line_number: Line number (1-indexed)
+        end_line: End line number (for multi-line symbols)
+        signature: Full signature (for functions/methods)
+        parent: Parent symbol name (for methods)
+        docstring: Documentation string if available
+    """
+    name: str
+    symbol_type: SymbolType
+    file_path: str
+    line_number: int
+    end_line: Optional[int] = None
+    signature: Optional[str] = None
+    parent: Optional[str] = None
+    docstring: Optional[str] = None
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary."""
+        return {
+            "name": self.name,
+            "type": self.symbol_type.value,
+            "file_path": self.file_path,
+            "line_number": self.line_number,
+            "end_line": self.end_line,
+            "signature": self.signature,
+            "parent": self.parent,
+            "docstring": self.docstring
+        }
+
+
+class LanguageParser:
+    """Base class for language-specific parsers."""
+    
+    def parse(self, content: str, file_path: str) -> List[Symbol]:
+        """Parse content and extract symbols.
+        
+        Args:
+            content: File content
+            file_path: Path to the file
+            
+        Returns:
+            List of extracted symbols
+        """
+        raise NotImplementedError
+
+
+class PythonParser(LanguageParser):
+    """Parser for Python files."""
+    
+    # Regex patterns for Python
+    FUNCTION_PATTERN = re.compile(
+        r'^(\s*)def\s+(\w+)\s*\((.*?)\)\s*(?:->.*?)?:',
+        re.MULTILINE
+    )
+    CLASS_PATTERN = re.compile(
+        r'^(\s*)class\s+(\w+)\s*(?:\((.*?)\))?:',
+        re.MULTILINE
+    )
+    IMPORT_PATTERN = re.compile(
+        r'^(?:from\s+(\S+)\s+)?import\s+(.+)$',
+        re.MULTILINE
+    )
+    ASYNC_FUNCTION_PATTERN = re.compile(
+        r'^(\s*)async\s+def\s+(\w+)\s*\((.*?)\)\s*(?:->.*?)?:',
+        re.MULTILINE
+    )
+    
+    def parse(self, content: str, file_path: str) -> List[Symbol]:
+        symbols = []
+        lines = content.split('\n')
+        
+        # Track class context for methods
+        current_class = None
+        class_indent = 0
+        
+        # Parse classes
+        for match in self.CLASS_PATTERN.finditer(content):
+            indent = len(match.group(1))
+            name = match.group(2)
+            bases = match.group(3) or ""
+            line_num = content[:match.start()].count('\n') + 1
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=SymbolType.CLASS,
+                file_path=file_path,
+                line_number=line_num,
+                signature=f"class {name}({bases})" if bases else f"class {name}"
+            ))
+        
+        # Parse functions (including async)
+        for pattern in [self.FUNCTION_PATTERN, self.ASYNC_FUNCTION_PATTERN]:
+            for match in pattern.finditer(content):
+                indent = len(match.group(1))
+                name = match.group(2)
+                params = match.group(3)
+                line_num = content[:match.start()].count('\n') + 1
+                
+                # Determine if it's a method (inside a class)
+                is_method = indent > 0
+                parent = None
+                
+                if is_method:
+                    # Find parent class
+                    for sym in symbols:
+                        if sym.symbol_type == SymbolType.CLASS:
+                            if sym.line_number < line_num:
+                                parent = sym.name
+                
+                symbol_type = SymbolType.METHOD if is_method else SymbolType.FUNCTION
+                is_async = "async" in match.group(0)
+                prefix = "async def" if is_async else "def"
+                
+                symbols.append(Symbol(
+                    name=name,
+                    symbol_type=symbol_type,
+                    file_path=file_path,
+                    line_number=line_num,
+                    signature=f"{prefix} {name}({params})",
+                    parent=parent
+                ))
+        
+        # Parse imports
+        for match in self.IMPORT_PATTERN.finditer(content):
+            from_module = match.group(1)
+            imports = match.group(2)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            if from_module:
+                sig = f"from {from_module} import {imports}"
+            else:
+                sig = f"import {imports}"
+            
+            symbols.append(Symbol(
+                name=imports.split(',')[0].strip().split(' ')[0],
+                symbol_type=SymbolType.IMPORT,
+                file_path=file_path,
+                line_number=line_num,
+                signature=sig
+            ))
+        
+        return symbols
+
+
+class JavaScriptParser(LanguageParser):
+    """Parser for JavaScript/TypeScript files."""
+    
+    FUNCTION_PATTERN = re.compile(
+        r'(?:export\s+)?(?:async\s+)?function\s+(\w+)\s*\((.*?)\)',
+        re.MULTILINE
+    )
+    ARROW_FUNCTION_PATTERN = re.compile(
+        r'(?:export\s+)?(?:const|let|var)\s+(\w+)\s*=\s*(?:async\s+)?\((.*?)\)\s*=>',
+        re.MULTILINE
+    )
+    CLASS_PATTERN = re.compile(
+        r'(?:export\s+)?class\s+(\w+)(?:\s+extends\s+(\w+))?',
+        re.MULTILINE
+    )
+    IMPORT_PATTERN = re.compile(
+        r'import\s+(?:{([^}]+)}|(\w+))\s+from\s+[\'"]([^\'"]+)[\'"]',
+        re.MULTILINE
+    )
+    EXPORT_PATTERN = re.compile(
+        r'export\s+(?:default\s+)?(?:const|let|var|function|class)\s+(\w+)',
+        re.MULTILINE
+    )
+    
+    def parse(self, content: str, file_path: str) -> List[Symbol]:
+        symbols = []
+        
+        # Parse functions
+        for match in self.FUNCTION_PATTERN.finditer(content):
+            name = match.group(1)
+            params = match.group(2)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=SymbolType.FUNCTION,
+                file_path=file_path,
+                line_number=line_num,
+                signature=f"function {name}({params})"
+            ))
+        
+        # Parse arrow functions
+        for match in self.ARROW_FUNCTION_PATTERN.finditer(content):
+            name = match.group(1)
+            params = match.group(2)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=SymbolType.FUNCTION,
+                file_path=file_path,
+                line_number=line_num,
+                signature=f"const {name} = ({params}) =>"
+            ))
+        
+        # Parse classes
+        for match in self.CLASS_PATTERN.finditer(content):
+            name = match.group(1)
+            extends = match.group(2)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            sig = f"class {name}"
+            if extends:
+                sig += f" extends {extends}"
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=SymbolType.CLASS,
+                file_path=file_path,
+                line_number=line_num,
+                signature=sig
+            ))
+        
+        # Parse imports
+        for match in self.IMPORT_PATTERN.finditer(content):
+            named = match.group(1)
+            default = match.group(2)
+            module = match.group(3)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            name = default or (named.split(',')[0].strip() if named else module)
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=SymbolType.IMPORT,
+                file_path=file_path,
+                line_number=line_num,
+                signature=f"import from '{module}'"
+            ))
+        
+        # Parse exports
+        for match in self.EXPORT_PATTERN.finditer(content):
+            name = match.group(1)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=SymbolType.EXPORT,
+                file_path=file_path,
+                line_number=line_num,
+                signature=f"export {name}"
+            ))
+        
+        return symbols
+
+
+class GoParser(LanguageParser):
+    """Parser for Go files."""
+    
+    FUNCTION_PATTERN = re.compile(
+        r'^func\s+(?:\((\w+)\s+\*?(\w+)\)\s+)?(\w+)\s*\((.*?)\)',
+        re.MULTILINE
+    )
+    TYPE_PATTERN = re.compile(
+        r'^type\s+(\w+)\s+(struct|interface)',
+        re.MULTILINE
+    )
+    IMPORT_PATTERN = re.compile(
+        r'import\s+(?:\(\s*([\s\S]*?)\s*\)|"([^"]+)")',
+        re.MULTILINE
+    )
+    
+    def parse(self, content: str, file_path: str) -> List[Symbol]:
+        symbols = []
+        
+        # Parse functions
+        for match in self.FUNCTION_PATTERN.finditer(content):
+            receiver_name = match.group(1)
+            receiver_type = match.group(2)
+            name = match.group(3)
+            params = match.group(4)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            if receiver_type:
+                # Method
+                symbols.append(Symbol(
+                    name=name,
+                    symbol_type=SymbolType.METHOD,
+                    file_path=file_path,
+                    line_number=line_num,
+                    signature=f"func ({receiver_name} {receiver_type}) {name}({params})",
+                    parent=receiver_type
+                ))
+            else:
+                # Function
+                symbols.append(Symbol(
+                    name=name,
+                    symbol_type=SymbolType.FUNCTION,
+                    file_path=file_path,
+                    line_number=line_num,
+                    signature=f"func {name}({params})"
+                ))
+        
+        # Parse types
+        for match in self.TYPE_PATTERN.finditer(content):
+            name = match.group(1)
+            kind = match.group(2)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            symbol_type = SymbolType.STRUCT if kind == "struct" else SymbolType.INTERFACE
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=symbol_type,
+                file_path=file_path,
+                line_number=line_num,
+                signature=f"type {name} {kind}"
+            ))
+        
+        return symbols
+
+
+class RustParser(LanguageParser):
+    """Parser for Rust files."""
+    
+    FUNCTION_PATTERN = re.compile(
+        r'^(\s*)(?:pub\s+)?(?:async\s+)?fn\s+(\w+)\s*(?:<[^>]*>)?\s*\((.*?)\)',
+        re.MULTILINE
+    )
+    STRUCT_PATTERN = re.compile(
+        r'^(?:pub\s+)?struct\s+(\w+)',
+        re.MULTILINE
+    )
+    IMPL_PATTERN = re.compile(
+        r'^impl(?:<[^>]*>)?\s+(?:(\w+)\s+for\s+)?(\w+)',
+        re.MULTILINE
+    )
+    USE_PATTERN = re.compile(
+        r'^use\s+(.+);',
+        re.MULTILINE
+    )
+    
+    def parse(self, content: str, file_path: str) -> List[Symbol]:
+        symbols = []
+        
+        # Parse functions
+        for match in self.FUNCTION_PATTERN.finditer(content):
+            indent = len(match.group(1))
+            name = match.group(2)
+            params = match.group(3)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            is_method = indent > 0
+            symbol_type = SymbolType.METHOD if is_method else SymbolType.FUNCTION
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=symbol_type,
+                file_path=file_path,
+                line_number=line_num,
+                signature=f"fn {name}({params})"
+            ))
+        
+        # Parse structs
+        for match in self.STRUCT_PATTERN.finditer(content):
+            name = match.group(1)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=SymbolType.STRUCT,
+                file_path=file_path,
+                line_number=line_num,
+                signature=f"struct {name}"
+            ))
+        
+        # Parse impl blocks
+        for match in self.IMPL_PATTERN.finditer(content):
+            trait_name = match.group(1)
+            type_name = match.group(2)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            if trait_name:
+                sig = f"impl {trait_name} for {type_name}"
+            else:
+                sig = f"impl {type_name}"
+            
+            symbols.append(Symbol(
+                name=type_name,
+                symbol_type=SymbolType.TYPE,
+                file_path=file_path,
+                line_number=line_num,
+                signature=sig
+            ))
+        
+        return symbols
+
+
+class JavaParser(LanguageParser):
+    """Parser for Java files."""
+    
+    CLASS_PATTERN = re.compile(
+        r'(?:public\s+)?(?:abstract\s+)?(?:final\s+)?class\s+(\w+)(?:\s+extends\s+(\w+))?(?:\s+implements\s+([^{]+))?',
+        re.MULTILINE
+    )
+    METHOD_PATTERN = re.compile(
+        r'(?:public|private|protected)?\s*(?:static\s+)?(?:final\s+)?(?:\w+(?:<[^>]+>)?)\s+(\w+)\s*\((.*?)\)',
+        re.MULTILINE
+    )
+    IMPORT_PATTERN = re.compile(
+        r'^import\s+(?:static\s+)?([^;]+);',
+        re.MULTILINE
+    )
+    
+    def parse(self, content: str, file_path: str) -> List[Symbol]:
+        symbols = []
+        
+        # Parse classes
+        for match in self.CLASS_PATTERN.finditer(content):
+            name = match.group(1)
+            extends = match.group(2)
+            implements = match.group(3)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            sig = f"class {name}"
+            if extends:
+                sig += f" extends {extends}"
+            if implements:
+                sig += f" implements {implements.strip()}"
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=SymbolType.CLASS,
+                file_path=file_path,
+                line_number=line_num,
+                signature=sig
+            ))
+        
+        # Parse methods
+        for match in self.METHOD_PATTERN.finditer(content):
+            name = match.group(1)
+            params = match.group(2)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            # Skip constructors and common keywords
+            if name in ('if', 'while', 'for', 'switch', 'catch'):
+                continue
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=SymbolType.METHOD,
+                file_path=file_path,
+                line_number=line_num,
+                signature=f"{name}({params})"
+            ))
+        
+        # Parse imports
+        for match in self.IMPORT_PATTERN.finditer(content):
+            import_path = match.group(1)
+            line_num = content[:match.start()].count('\n') + 1
+            
+            # Get the last part of the import
+            name = import_path.split('.')[-1]
+            
+            symbols.append(Symbol(
+                name=name,
+                symbol_type=SymbolType.IMPORT,
+                file_path=file_path,
+                line_number=line_num,
+                signature=f"import {import_path}"
+            ))
+        
+        return symbols
+
+
+class SymbolIndexer:
+    """Indexes code symbols from source files.
+    
+    Supports multiple languages:
+    - Python (.py)
+    - JavaScript (.js, .jsx)
+    - TypeScript (.ts, .tsx)
+    - Go (.go)
+    - Rust (.rs)
+    - Java (.java)
+    """
+    
+    # Map file extensions to parsers
+    PARSERS = {
+        '.py': PythonParser,
+        '.js': JavaScriptParser,
+        '.jsx': JavaScriptParser,
+        '.ts': JavaScriptParser,
+        '.tsx': JavaScriptParser,
+        '.go': GoParser,
+        '.rs': RustParser,
+        '.java': JavaParser,
+    }
+    
+    def __init__(self, workspace_path: str):
+        """Initialize symbol indexer.
+        
+        Args:
+            workspace_path: Root directory to index
+        """
+        self.workspace_path = os.path.abspath(workspace_path)
+        self.symbols: Dict[str, List[Symbol]] = {}  # file_path -> symbols
+        self.by_name: Dict[str, List[Symbol]] = {}  # symbol_name -> symbols
+        self.by_type: Dict[SymbolType, List[Symbol]] = {}  # type -> symbols
+        
+        # Initialize parsers
+        self._parsers = {ext: parser() for ext, parser in self.PARSERS.items()}
+    
+    def index_file(self, file_path: str) -> List[Symbol]:
+        """Index symbols from a single file.
+        
+        Args:
+            file_path: Path to the file
+            
+        Returns:
+            List of extracted symbols
+        """
+        ext = os.path.splitext(file_path)[1].lower()
+        parser = self._parsers.get(ext)
+        
+        if not parser:
+            return []
+        
+        try:
+            with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
+                content = f.read()
+            
+            rel_path = os.path.relpath(file_path, self.workspace_path)
+            symbols = parser.parse(content, rel_path)
+            
+            # Store in indexes
+            self.symbols[rel_path] = symbols
+            
+            for sym in symbols:
+                # By name
+                if sym.name not in self.by_name:
+                    self.by_name[sym.name] = []
+                self.by_name[sym.name].append(sym)
+                
+                # By type
+                if sym.symbol_type not in self.by_type:
+                    self.by_type[sym.symbol_type] = []
+                self.by_type[sym.symbol_type].append(sym)
+            
+            return symbols
+        except Exception as e:
+            logger.debug(f"Error indexing {file_path}: {e}")
+            return []
+    
+    def index(self) -> int:
+        """Index all supported files in workspace.
+        
+        Returns:
+            Total number of symbols indexed
+        """
+        total_symbols = 0
+        
+        for root, dirs, files in os.walk(self.workspace_path):
+            # Skip hidden directories
+            dirs[:] = [d for d in dirs if not d.startswith('.')]
+            
+            for filename in files:
+                ext = os.path.splitext(filename)[1].lower()
+                if ext in self.PARSERS:
+                    file_path = os.path.join(root, filename)
+                    symbols = self.index_file(file_path)
+                    total_symbols += len(symbols)
+        
+        logger.info(f"Indexed {total_symbols} symbols from {len(self.symbols)} files")
+        return total_symbols
+    
+    def find_by_name(self, name: str, exact: bool = False) -> List[Symbol]:
+        """Find symbols by name.
+        
+        Args:
+            name: Symbol name to search
+            exact: If True, match exact name
+            
+        Returns:
+            List of matching symbols
+        """
+        if exact:
+            return self.by_name.get(name, [])
+        
+        results = []
+        name_lower = name.lower()
+        for sym_name, symbols in self.by_name.items():
+            if name_lower in sym_name.lower():
+                results.extend(symbols)
+        return results
+    
+    def find_by_type(self, symbol_type: SymbolType) -> List[Symbol]:
+        """Find symbols by type.
+        
+        Args:
+            symbol_type: Type of symbol
+            
+        Returns:
+            List of matching symbols
+        """
+        return self.by_type.get(symbol_type, [])
+    
+    def find_in_file(self, file_path: str) -> List[Symbol]:
+        """Find all symbols in a file.
+        
+        Args:
+            file_path: Path to the file
+            
+        Returns:
+            List of symbols in the file
+        """
+        return self.symbols.get(file_path, [])
+    
+    def get_stats(self) -> Dict[str, Any]:
+        """Get indexer statistics.
+        
+        Returns:
+            Dictionary with stats
+        """
+        type_counts = {t.value: len(syms) for t, syms in self.by_type.items()}
+        
+        return {
+            "total_files": len(self.symbols),
+            "total_symbols": sum(len(s) for s in self.symbols.values()),
+            "by_type": type_counts
+        }

src/praisonai-agents/praisonaiagents/context/fast/parallel_executor.py

@@ -0,0 +1,325 @@
+"""
+Parallel execution engine for Fast Context.
+
+Provides async parallel execution of search tools with:
+- Configurable parallelism (default: 8 concurrent calls)
+- Timeout handling per tool call
+- Result aggregation
+- Error isolation (one failure doesn't affect others)
+"""
+
+import asyncio
+from concurrent.futures import ThreadPoolExecutor
+from typing import List, Dict, Any, Optional, Callable
+import logging
+import time
+
+logger = logging.getLogger(__name__)
+
+
+class ToolCallBatch:
+    """Batch of tool calls to execute in parallel.
+    
+    Attributes:
+        tasks: List of task dictionaries
+        max_size: Maximum batch size (default: 8)
+    """
+    
+    def __init__(self, max_size: int = 8):
+        """Initialize batch.
+        
+        Args:
+            max_size: Maximum number of tasks in batch
+        """
+        self.tasks: List[Dict[str, Any]] = []
+        self.max_size = max_size
+    
+    def add(self, tool_name: str, **kwargs) -> bool:
+        """Add a tool call to the batch.
+        
+        Args:
+            tool_name: Name of the tool to call
+            **kwargs: Tool arguments
+            
+        Returns:
+            True if added, False if batch is full
+        """
+        if self.is_full:
+            return False
+        
+        self.tasks.append({
+            "tool": tool_name,
+            "args": kwargs
+        })
+        return True
+    
+    @property
+    def is_full(self) -> bool:
+        """Check if batch is at max capacity."""
+        return len(self.tasks) >= self.max_size
+    
+    def clear(self) -> None:
+        """Clear all tasks from batch."""
+        self.tasks.clear()
+
+
+class ParallelExecutor:
+    """Executes search tools in parallel.
+    
+    Uses asyncio for concurrent execution with a thread pool
+    for CPU-bound operations.
+    
+    Attributes:
+        max_parallel: Maximum concurrent executions
+        timeout: Timeout per tool call in seconds
+    """
+    
+    def __init__(
+        self,
+        max_parallel: int = 8,
+        timeout: float = 30.0
+    ):
+        """Initialize executor.
+        
+        Args:
+            max_parallel: Maximum concurrent tool calls
+            timeout: Timeout per tool call in seconds
+        """
+        self.max_parallel = max_parallel
+        self.timeout = timeout
+        self._executor: Optional[ThreadPoolExecutor] = None
+    
+    def _get_executor(self) -> ThreadPoolExecutor:
+        """Get or create thread pool executor."""
+        if self._executor is None:
+            self._executor = ThreadPoolExecutor(max_workers=self.max_parallel)
+        return self._executor
+    
+    def _get_tool_function(self, tool_name: str) -> Optional[Callable]:
+        """Get tool function by name.
+        
+        Args:
+            tool_name: Name of the tool
+            
+        Returns:
+            Tool function or None if not found
+        """
+        # Import here to avoid circular imports
+        from praisonaiagents.context.fast.search_tools import (
+            grep_search,
+            glob_search,
+            read_file,
+            list_directory
+        )
+        
+        tools = {
+            "grep_search": grep_search,
+            "glob_search": glob_search,
+            "read_file": read_file,
+            "list_directory": list_directory
+        }
+        
+        return tools.get(tool_name)
+    
+    async def _execute_single(
+        self,
+        task: Dict[str, Any],
+        loop: asyncio.AbstractEventLoop
+    ) -> Any:
+        """Execute a single tool call.
+        
+        Args:
+            task: Task dictionary with 'tool' and 'args'
+            loop: Event loop for thread execution
+            
+        Returns:
+            Tool result or error dictionary
+        """
+        tool_name = task.get("tool")
+        args = task.get("args", {})
+        
+        tool_func = self._get_tool_function(tool_name)
+        if tool_func is None:
+            return {"error": f"Unknown tool: {tool_name}"}
+        
+        try:
+            # Run in thread pool to avoid blocking
+            executor = self._get_executor()
+            result = await asyncio.wait_for(
+                loop.run_in_executor(executor, lambda: tool_func(**args)),
+                timeout=self.timeout
+            )
+            return result
+        except asyncio.TimeoutError:
+            return {"error": f"Timeout executing {tool_name}", "timeout": True}
+        except Exception as e:
+            logger.debug(f"Error executing {tool_name}: {e}")
+            return {"error": str(e)}
+    
+    async def execute(self, tasks: List[Dict[str, Any]]) -> List[Any]:
+        """Execute multiple tool calls in parallel.
+        
+        Args:
+            tasks: List of task dictionaries, each with 'tool' and 'args'
+            
+        Returns:
+            List of results in same order as tasks
+        """
+        if not tasks:
+            return []
+        
+        loop = asyncio.get_event_loop()
+        
+        # Create semaphore to limit concurrency
+        semaphore = asyncio.Semaphore(self.max_parallel)
+        
+        async def bounded_execute(task: Dict[str, Any]) -> Any:
+            async with semaphore:
+                return await self._execute_single(task, loop)
+        
+        # Execute all tasks concurrently
+        results = await asyncio.gather(
+            *[bounded_execute(task) for task in tasks],
+            return_exceptions=True
+        )
+        
+        # Convert exceptions to error dicts
+        processed_results = []
+        for result in results:
+            if isinstance(result, Exception):
+                processed_results.append({"error": str(result)})
+            else:
+                processed_results.append(result)
+        
+        return processed_results
+    
+    def execute_sync(self, tasks: List[Dict[str, Any]]) -> List[Any]:
+        """Synchronous wrapper for execute.
+        
+        Args:
+            tasks: List of task dictionaries
+            
+        Returns:
+            List of results
+        """
+        try:
+            loop = asyncio.get_event_loop()
+            if loop.is_running():
+                # If already in async context, create new loop
+                import concurrent.futures
+                with concurrent.futures.ThreadPoolExecutor() as pool:
+                    future = pool.submit(asyncio.run, self.execute(tasks))
+                    return future.result()
+            else:
+                return loop.run_until_complete(self.execute(tasks))
+        except RuntimeError:
+            # No event loop, create one
+            return asyncio.run(self.execute(tasks))
+    
+    def close(self) -> None:
+        """Close the executor and release resources."""
+        if self._executor is not None:
+            self._executor.shutdown(wait=False)
+            self._executor = None
+    
+    def __enter__(self):
+        return self
+    
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self.close()
+
+
+class ParallelSearchCoordinator:
+    """Coordinates parallel search operations across multiple turns.
+    
+    Implements the Fast Context search strategy:
+    - Up to 8 parallel tool calls per turn
+    - Maximum 4 turns
+    - Aggregates results across turns
+    """
+    
+    def __init__(
+        self,
+        max_parallel: int = 8,
+        max_turns: int = 4,
+        timeout: float = 30.0
+    ):
+        """Initialize coordinator.
+        
+        Args:
+            max_parallel: Max parallel calls per turn
+            max_turns: Maximum number of turns
+            timeout: Timeout per tool call
+        """
+        self.max_parallel = max_parallel
+        self.max_turns = max_turns
+        self.timeout = timeout
+        self.executor = ParallelExecutor(max_parallel, timeout)
+        
+        # Track execution stats
+        self.total_tool_calls = 0
+        self.turns_used = 0
+        self.start_time: Optional[float] = None
+        self.end_time: Optional[float] = None
+    
+    def start(self) -> None:
+        """Start timing the search."""
+        self.start_time = time.perf_counter()
+        self.total_tool_calls = 0
+        self.turns_used = 0
+    
+    def stop(self) -> None:
+        """Stop timing the search."""
+        self.end_time = time.perf_counter()
+    
+    @property
+    def elapsed_ms(self) -> int:
+        """Get elapsed time in milliseconds."""
+        if self.start_time is None:
+            return 0
+        end = self.end_time or time.perf_counter()
+        return int((end - self.start_time) * 1000)
+    
+    async def execute_turn(self, batch: ToolCallBatch) -> List[Any]:
+        """Execute a single turn of parallel tool calls.
+        
+        Args:
+            batch: Batch of tool calls
+            
+        Returns:
+            List of results
+        """
+        if self.turns_used >= self.max_turns:
+            logger.warning("Maximum turns reached")
+            return []
+        
+        self.turns_used += 1
+        self.total_tool_calls += len(batch.tasks)
+        
+        results = await self.executor.execute(batch.tasks)
+        return results
+    
+    def execute_turn_sync(self, batch: ToolCallBatch) -> List[Any]:
+        """Synchronous wrapper for execute_turn.
+        
+        Args:
+            batch: Batch of tool calls
+            
+        Returns:
+            List of results
+        """
+        try:
+            loop = asyncio.get_event_loop()
+            if loop.is_running():
+                import concurrent.futures
+                with concurrent.futures.ThreadPoolExecutor() as pool:
+                    future = pool.submit(asyncio.run, self.execute_turn(batch))
+                    return future.result()
+            else:
+                return loop.run_until_complete(self.execute_turn(batch))
+        except RuntimeError:
+            return asyncio.run(self.execute_turn(batch))
+    
+    def close(self) -> None:
+        """Close resources."""
+        self.executor.close()

src/praisonai-agents/praisonaiagents/context/fast/result.py

@@ -0,0 +1,210 @@
+"""
+Result classes for Fast Context.
+
+These dataclasses represent the output of Fast Context searches,
+providing structured file and line range information.
+"""
+
+from dataclasses import dataclass, field
+from typing import List, Optional
+import time
+
+
+@dataclass
+class LineRange:
+    """Represents a range of lines in a file.
+    
+    Attributes:
+        start: Starting line number (1-indexed)
+        end: Ending line number (1-indexed, inclusive)
+        content: Optional content of the lines
+        relevance_score: Score indicating relevance (0.0-1.0)
+    """
+    start: int
+    end: int
+    content: Optional[str] = None
+    relevance_score: float = 1.0
+    
+    def __post_init__(self):
+        if self.start < 1:
+            self.start = 1
+        if self.end < self.start:
+            self.end = self.start
+    
+    @property
+    def line_count(self) -> int:
+        """Number of lines in this range."""
+        return self.end - self.start + 1
+    
+    def overlaps(self, other: "LineRange") -> bool:
+        """Check if this range overlaps with another."""
+        return not (self.end < other.start or self.start > other.end)
+    
+    def merge(self, other: "LineRange") -> "LineRange":
+        """Merge with another overlapping range."""
+        if not self.overlaps(other):
+            raise ValueError("Cannot merge non-overlapping ranges")
+        return LineRange(
+            start=min(self.start, other.start),
+            end=max(self.end, other.end),
+            content=None,  # Content needs to be re-fetched
+            relevance_score=max(self.relevance_score, other.relevance_score)
+        )
+
+
+@dataclass
+class FileMatch:
+    """Represents a file match from Fast Context search.
+    
+    Attributes:
+        path: Absolute or relative path to the file
+        line_ranges: List of relevant line ranges in the file
+        relevance_score: Overall relevance score (0.0-1.0)
+        match_count: Number of pattern matches found
+    """
+    path: str
+    line_ranges: List[LineRange] = field(default_factory=list)
+    relevance_score: float = 1.0
+    match_count: int = 0
+    
+    def add_line_range(self, line_range: LineRange) -> None:
+        """Add a line range, merging with existing overlapping ranges."""
+        # Find overlapping ranges
+        overlapping = []
+        non_overlapping = []
+        for existing in self.line_ranges:
+            if existing.overlaps(line_range):
+                overlapping.append(existing)
+            else:
+                non_overlapping.append(existing)
+        
+        # Merge all overlapping ranges
+        merged = line_range
+        for overlap in overlapping:
+            merged = merged.merge(overlap)
+        
+        self.line_ranges = non_overlapping + [merged]
+        self.line_ranges.sort(key=lambda r: r.start)
+    
+    @property
+    def total_lines(self) -> int:
+        """Total number of lines across all ranges."""
+        return sum(r.line_count for r in self.line_ranges)
+
+
+@dataclass
+class FastContextResult:
+    """Result of a Fast Context search operation.
+    
+    Attributes:
+        files: List of matching files with line ranges
+        query: The original search query
+        search_time_ms: Time taken for the search in milliseconds
+        turns_used: Number of turns used by the subagent
+        total_tool_calls: Total number of tool calls made
+        from_cache: Whether result was retrieved from cache
+    """
+    files: List[FileMatch] = field(default_factory=list)
+    query: str = ""
+    search_time_ms: int = 0
+    turns_used: int = 0
+    total_tool_calls: int = 0
+    from_cache: bool = False
+    
+    _start_time: float = field(default=0.0, repr=False)
+    
+    def start_timer(self) -> None:
+        """Start the search timer."""
+        self._start_time = time.perf_counter()
+    
+    def stop_timer(self) -> None:
+        """Stop the timer and record elapsed time."""
+        if self._start_time > 0:
+            elapsed = time.perf_counter() - self._start_time
+            self.search_time_ms = int(elapsed * 1000)
+    
+    def add_file(self, file_match: FileMatch) -> None:
+        """Add a file match, merging with existing if same path."""
+        for existing in self.files:
+            if existing.path == file_match.path:
+                # Merge line ranges
+                for lr in file_match.line_ranges:
+                    existing.add_line_range(lr)
+                existing.relevance_score = max(
+                    existing.relevance_score, 
+                    file_match.relevance_score
+                )
+                existing.match_count += file_match.match_count
+                return
+        self.files.append(file_match)
+    
+    def sort_by_relevance(self) -> None:
+        """Sort files by relevance score (highest first)."""
+        self.files.sort(key=lambda f: f.relevance_score, reverse=True)
+    
+    @property
+    def total_files(self) -> int:
+        """Total number of files found."""
+        return len(self.files)
+    
+    @property
+    def total_lines(self) -> int:
+        """Total number of lines across all files."""
+        return sum(f.total_lines for f in self.files)
+    
+    def to_context_string(self, max_files: int = 10, include_content: bool = True) -> str:
+        """Convert result to a context string for the main agent.
+        
+        Args:
+            max_files: Maximum number of files to include
+            include_content: Whether to include line content
+            
+        Returns:
+            Formatted context string
+        """
+        if not self.files:
+            return "No relevant files found."
+        
+        lines = [f"Found {self.total_files} relevant file(s):"]
+        
+        for i, file_match in enumerate(self.files[:max_files]):
+            lines.append(f"\n## {file_match.path}")
+            if file_match.line_ranges:
+                for lr in file_match.line_ranges:
+                    lines.append(f"  Lines {lr.start}-{lr.end}")
+                    if include_content and lr.content:
+                        # Indent content
+                        content_lines = lr.content.split('\n')
+                        for cl in content_lines:
+                            lines.append(f"    {cl}")
+        
+        if len(self.files) > max_files:
+            lines.append(f"\n... and {len(self.files) - max_files} more files")
+        
+        return '\n'.join(lines)
+    
+    def to_dict(self) -> dict:
+        """Convert to dictionary for serialization."""
+        return {
+            "files": [
+                {
+                    "path": f.path,
+                    "relevance_score": f.relevance_score,
+                    "match_count": f.match_count,
+                    "line_ranges": [
+                        {
+                            "start": lr.start,
+                            "end": lr.end,
+                            "relevance_score": lr.relevance_score
+                        }
+                        for lr in f.line_ranges
+                    ]
+                }
+                for f in self.files
+            ],
+            "query": self.query,
+            "search_time_ms": self.search_time_ms,
+            "turns_used": self.turns_used,
+            "total_tool_calls": self.total_tool_calls,
+            "from_cache": self.from_cache
+        }

src/praisonai-agents/praisonaiagents/context/fast/search_strategy.py

@@ -0,0 +1,368 @@
+"""
+Search Strategy for Fast Context.
+
+Orchestrates multi-turn search with:
+- Turn 1-3: Exploration (grep, glob, directory listing)
+- Turn 4: Final answer (file list + line ranges)
+- Early termination if sufficient context found
+"""
+
+import logging
+from dataclasses import dataclass, field
+from typing import List, Dict, Any, Optional, Set
+from enum import Enum
+
+from praisonaiagents.context.fast.result import FastContextResult, FileMatch, LineRange
+from praisonaiagents.context.fast.parallel_executor import ToolCallBatch
+
+logger = logging.getLogger(__name__)
+
+
+class SearchPhase(Enum):
+    """Phases of the search strategy."""
+    DISCOVERY = "discovery"      # Turn 1: Discover structure
+    EXPLORATION = "exploration"  # Turn 2-3: Deep search
+    REFINEMENT = "refinement"    # Turn 4: Final answer
+
+
+@dataclass
+class SearchState:
+    """Tracks the state of a multi-turn search.
+    
+    Attributes:
+        query: Original search query
+        current_turn: Current turn number (1-indexed)
+        max_turns: Maximum turns allowed
+        phase: Current search phase
+        discovered_files: Files found so far
+        explored_paths: Paths already explored
+        result: Accumulated search result
+        should_terminate: Whether to terminate early
+    """
+    query: str
+    current_turn: int = 0
+    max_turns: int = 4
+    phase: SearchPhase = SearchPhase.DISCOVERY
+    discovered_files: Set[str] = field(default_factory=set)
+    explored_paths: Set[str] = field(default_factory=set)
+    result: FastContextResult = field(default_factory=FastContextResult)
+    should_terminate: bool = False
+    
+    # Thresholds for early termination
+    min_files_for_termination: int = 5
+    max_files_before_refinement: int = 20
+    
+    def advance_turn(self) -> None:
+        """Advance to the next turn and update phase."""
+        self.current_turn += 1
+        
+        if self.current_turn == 1:
+            self.phase = SearchPhase.DISCOVERY
+        elif self.current_turn < self.max_turns:
+            self.phase = SearchPhase.EXPLORATION
+        else:
+            self.phase = SearchPhase.REFINEMENT
+    
+    def check_early_termination(self) -> bool:
+        """Check if search should terminate early.
+        
+        Returns:
+            True if search should terminate
+        """
+        # Terminate if we have enough high-quality results
+        if len(self.discovered_files) >= self.min_files_for_termination:
+            high_relevance = sum(
+                1 for f in self.result.files 
+                if f.relevance_score >= 0.8
+            )
+            if high_relevance >= 3:
+                self.should_terminate = True
+                return True
+        
+        # Terminate if we've explored everything
+        if self.current_turn >= 2 and not self._has_unexplored_paths():
+            self.should_terminate = True
+            return True
+        
+        return False
+    
+    def _has_unexplored_paths(self) -> bool:
+        """Check if there are unexplored paths."""
+        # Simple heuristic: if we found files but haven't explored many paths
+        return len(self.discovered_files) > len(self.explored_paths) * 2
+
+
+class SearchStrategy:
+    """Orchestrates multi-turn search strategy.
+    
+    Implements the Fast Context search approach:
+    - Turn 1: Discovery - understand codebase structure
+    - Turn 2-3: Exploration - deep search for relevant code
+    - Turn 4: Refinement - finalize results
+    
+    Supports early termination when sufficient context is found.
+    """
+    
+    def __init__(
+        self,
+        workspace_path: str,
+        max_turns: int = 4,
+        max_parallel: int = 8
+    ):
+        """Initialize search strategy.
+        
+        Args:
+            workspace_path: Root directory for searches
+            max_turns: Maximum search turns
+            max_parallel: Maximum parallel tool calls per turn
+        """
+        self.workspace_path = workspace_path
+        self.max_turns = max_turns
+        self.max_parallel = max_parallel
+    
+    def create_state(self, query: str) -> SearchState:
+        """Create a new search state.
+        
+        Args:
+            query: Search query
+            
+        Returns:
+            New SearchState instance
+        """
+        state = SearchState(query=query, max_turns=self.max_turns)
+        state.result.query = query
+        return state
+    
+    def plan_discovery_turn(self, state: SearchState) -> ToolCallBatch:
+        """Plan the discovery turn (Turn 1).
+        
+        Focus on understanding codebase structure:
+        - List root directory
+        - Find common file patterns
+        - Initial grep for query terms
+        
+        Args:
+            state: Current search state
+            
+        Returns:
+            Batch of tool calls to execute
+        """
+        batch = ToolCallBatch(max_size=self.max_parallel)
+        query = state.query
+        
+        # List root directory
+        batch.add("list_directory", dir_path=self.workspace_path, recursive=False)
+        
+        # Find Python files (most common)
+        batch.add("glob_search", search_path=self.workspace_path, pattern="**/*.py", max_results=20)
+        
+        # Initial grep for query terms
+        query_words = query.lower().split()
+        for word in query_words[:3]:  # First 3 words
+            if len(word) >= 3:  # Skip short words
+                batch.add("grep_search", search_path=self.workspace_path, pattern=word, max_results=10)
+                if batch.is_full:
+                    break
+        
+        return batch
+    
+    def plan_exploration_turn(self, state: SearchState) -> ToolCallBatch:
+        """Plan an exploration turn (Turn 2-3).
+        
+        Deep search based on discovery results:
+        - Search in discovered directories
+        - Read promising files
+        - Follow references
+        
+        Args:
+            state: Current search state
+            
+        Returns:
+            Batch of tool calls to execute
+        """
+        batch = ToolCallBatch(max_size=self.max_parallel)
+        
+        # Read top files found so far
+        for file_match in state.result.files[:4]:
+            if file_match.path not in state.explored_paths:
+                batch.add("read_file", filepath=file_match.path, context_lines=5)
+                state.explored_paths.add(file_match.path)
+                if batch.is_full:
+                    break
+        
+        # Search in unexplored directories
+        unexplored_dirs = self._find_unexplored_dirs(state)
+        for dir_path in unexplored_dirs[:2]:
+            batch.add("list_directory", dir_path=dir_path, recursive=True, max_depth=2)
+            if batch.is_full:
+                break
+        
+        # Additional grep searches with refined patterns
+        if not batch.is_full:
+            refined_patterns = self._generate_refined_patterns(state)
+            for pattern in refined_patterns[:2]:
+                batch.add("grep_search", search_path=self.workspace_path, pattern=pattern, max_results=15)
+                if batch.is_full:
+                    break
+        
+        return batch
+    
+    def plan_refinement_turn(self, state: SearchState) -> ToolCallBatch:
+        """Plan the refinement turn (Turn 4).
+        
+        Finalize results:
+        - Read remaining important files
+        - Get specific line ranges
+        
+        Args:
+            state: Current search state
+            
+        Returns:
+            Batch of tool calls to execute
+        """
+        batch = ToolCallBatch(max_size=self.max_parallel)
+        
+        # Read top files that haven't been fully explored
+        for file_match in state.result.files[:self.max_parallel]:
+            if file_match.path not in state.explored_paths:
+                # Read specific line ranges if available
+                if file_match.line_ranges:
+                    for lr in file_match.line_ranges[:2]:
+                        batch.add(
+                            "read_file",
+                            filepath=file_match.path,
+                            start_line=max(1, lr.start - 5),
+                            end_line=lr.end + 5
+                        )
+                        if batch.is_full:
+                            break
+                else:
+                    batch.add("read_file", filepath=file_match.path)
+                
+                state.explored_paths.add(file_match.path)
+                if batch.is_full:
+                    break
+        
+        return batch
+    
+    def plan_next_turn(self, state: SearchState) -> Optional[ToolCallBatch]:
+        """Plan the next turn based on current state.
+        
+        Args:
+            state: Current search state
+            
+        Returns:
+            Batch of tool calls, or None if search should end
+        """
+        state.advance_turn()
+        
+        # Check for early termination
+        if state.should_terminate or state.current_turn > state.max_turns:
+            return None
+        
+        if state.check_early_termination():
+            return None
+        
+        # Plan based on phase
+        if state.phase == SearchPhase.DISCOVERY:
+            return self.plan_discovery_turn(state)
+        elif state.phase == SearchPhase.EXPLORATION:
+            return self.plan_exploration_turn(state)
+        else:  # REFINEMENT
+            return self.plan_refinement_turn(state)
+    
+    def process_results(
+        self,
+        state: SearchState,
+        tool_results: List[Any]
+    ) -> None:
+        """Process results from a turn and update state.
+        
+        Args:
+            state: Current search state
+            tool_results: Results from tool execution
+        """
+        for result in tool_results:
+            if isinstance(result, list):
+                # Grep or glob results
+                for item in result:
+                    if isinstance(item, dict):
+                        path = item.get("path", "")
+                        if path:
+                            state.discovered_files.add(path)
+                            
+                            # Create or update file match
+                            file_match = FileMatch(
+                                path=path,
+                                relevance_score=0.8,
+                                match_count=1
+                            )
+                            
+                            # Add line range if available
+                            line_num = item.get("line_number")
+                            if line_num:
+                                file_match.add_line_range(LineRange(
+                                    start=max(1, line_num - 2),
+                                    end=line_num + 2,
+                                    content=item.get("context"),
+                                    relevance_score=0.8
+                                ))
+                            
+                            state.result.add_file(file_match)
+            
+            elif isinstance(result, dict):
+                # Read file or list directory result
+                if result.get("success"):
+                    path = result.get("path", "")
+                    if path:
+                        state.explored_paths.add(path)
+                    
+                    # Handle directory listing
+                    entries = result.get("entries", [])
+                    for entry in entries:
+                        if not entry.get("is_dir"):
+                            state.discovered_files.add(entry.get("path", ""))
+    
+    def _find_unexplored_dirs(self, state: SearchState) -> List[str]:
+        """Find directories that haven't been explored.
+        
+        Args:
+            state: Current search state
+            
+        Returns:
+            List of unexplored directory paths
+        """
+        import os
+        
+        # Extract unique directories from discovered files
+        dirs = set()
+        for file_path in state.discovered_files:
+            dir_path = os.path.dirname(file_path)
+            if dir_path and dir_path not in state.explored_paths:
+                dirs.add(dir_path)
+        
+        return list(dirs)[:5]  # Limit to 5
+    
+    def _generate_refined_patterns(self, state: SearchState) -> List[str]:
+        """Generate refined search patterns based on results.
+        
+        Args:
+            state: Current search state
+            
+        Returns:
+            List of refined search patterns
+        """
+        patterns = []
+        query_words = state.query.lower().split()
+        
+        # Combine query words
+        if len(query_words) >= 2:
+            patterns.append(f"{query_words[0]}.*{query_words[1]}")
+        
+        # Add common code patterns
+        for word in query_words[:2]:
+            if len(word) >= 3:
+                patterns.append(f"def {word}")
+                patterns.append(f"class {word}")
+        
+        return patterns[:4]

src/praisonai-agents/praisonaiagents/context/fast/search_tools.py

@@ -0,0 +1,636 @@
+"""
+Search tools for Fast Context.
+
+Provides cross-platform compatible search tools:
+- grep_search: Pattern search in files
+- glob_search: File pattern matching
+- read_file: Read file contents with line range support
+- list_directory: List directory contents
+
+These tools are designed to be fast, safe, and cross-platform compatible.
+"""
+
+import os
+import re
+import fnmatch
+from pathlib import Path
+from typing import List, Dict, Any, Optional, Set
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+def _load_gitignore_patterns(root_path: str) -> Set[str]:
+    """Load patterns from .gitignore file.
+    
+    Args:
+        root_path: Root directory to search for .gitignore
+        
+    Returns:
+        Set of gitignore patterns
+    """
+    patterns = set()
+    gitignore_path = Path(root_path) / ".gitignore"
+    
+    if gitignore_path.exists():
+        try:
+            with open(gitignore_path, 'r', encoding='utf-8', errors='ignore') as f:
+                for line in f:
+                    line = line.strip()
+                    # Skip empty lines and comments
+                    if line and not line.startswith('#'):
+                        patterns.add(line)
+        except Exception:
+            pass
+    
+    # Also check for .praisonignore
+    praisonignore_path = Path(root_path) / ".praisonignore"
+    if praisonignore_path.exists():
+        try:
+            with open(praisonignore_path, 'r', encoding='utf-8', errors='ignore') as f:
+                for line in f:
+                    line = line.strip()
+                    if line and not line.startswith('#'):
+                        patterns.add(line)
+        except Exception:
+            pass
+    
+    return patterns
+
+
+def _should_ignore(path: str, root_path: str, patterns: Set[str]) -> bool:
+    """Check if a path should be ignored based on gitignore patterns.
+    
+    Args:
+        path: Path to check
+        root_path: Root directory
+        patterns: Set of gitignore patterns
+        
+    Returns:
+        True if path should be ignored
+    """
+    if not patterns:
+        return False
+    
+    rel_path = os.path.relpath(path, root_path)
+    path_parts = Path(rel_path).parts
+    
+    for pattern in patterns:
+        # Handle directory patterns (ending with /)
+        if pattern.endswith('/'):
+            dir_pattern = pattern.rstrip('/')
+            if any(fnmatch.fnmatch(part, dir_pattern) for part in path_parts):
+                return True
+        # Handle ** patterns
+        elif '**' in pattern:
+            # Convert gitignore pattern to fnmatch pattern
+            fnmatch_pattern = pattern.replace('**/', '*').replace('**', '*')
+            if fnmatch.fnmatch(rel_path, fnmatch_pattern):
+                return True
+        # Handle simple patterns
+        else:
+            # Check against filename
+            if fnmatch.fnmatch(os.path.basename(path), pattern):
+                return True
+            # Check against relative path
+            if fnmatch.fnmatch(rel_path, pattern):
+                return True
+            # Check if any path component matches
+            if any(fnmatch.fnmatch(part, pattern) for part in path_parts):
+                return True
+    
+    return False
+
+
+def _is_binary_file(filepath: str) -> bool:
+    """Check if a file is binary.
+    
+    Args:
+        filepath: Path to the file
+        
+    Returns:
+        True if file appears to be binary
+    """
+    try:
+        with open(filepath, 'rb') as f:
+            chunk = f.read(8192)
+            # Check for null bytes (common in binary files)
+            if b'\x00' in chunk:
+                return True
+            # Check for high ratio of non-text characters
+            text_chars = bytearray({7, 8, 9, 10, 12, 13, 27} | set(range(0x20, 0x100)) - {0x7f})
+            non_text = sum(1 for byte in chunk if byte not in text_chars)
+            if len(chunk) > 0 and non_text / len(chunk) > 0.3:
+                return True
+    except Exception:
+        return True
+    return False
+
+
+def grep_search(
+    search_path: str,
+    pattern: str,
+    is_regex: bool = False,
+    case_sensitive: bool = False,
+    include_patterns: Optional[List[str]] = None,
+    exclude_patterns: Optional[List[str]] = None,
+    max_results: int = 100,
+    context_lines: int = 0,
+    respect_gitignore: bool = True
+) -> List[Dict[str, Any]]:
+    """Search for pattern in files.
+    
+    Args:
+        search_path: Directory or file to search
+        pattern: Search pattern (string or regex)
+        is_regex: If True, treat pattern as regex
+        case_sensitive: If True, search is case sensitive
+        include_patterns: Glob patterns to include (e.g., ["*.py"])
+        exclude_patterns: Glob patterns to exclude (e.g., ["**/test_*"])
+        max_results: Maximum number of results to return
+        context_lines: Number of context lines before/after match
+        respect_gitignore: If True, respect .gitignore patterns
+        
+    Returns:
+        List of match dictionaries with path, line_number, content, context
+    """
+    results = []
+    search_path = os.path.abspath(search_path)
+    
+    # Compile pattern
+    if is_regex:
+        flags = 0 if case_sensitive else re.IGNORECASE
+        try:
+            regex = re.compile(pattern, flags)
+        except re.error as e:
+            logger.warning(f"Invalid regex pattern: {e}")
+            return []
+    else:
+        if case_sensitive:
+            regex = re.compile(re.escape(pattern))
+        else:
+            regex = re.compile(re.escape(pattern), re.IGNORECASE)
+    
+    # Load gitignore patterns
+    gitignore_patterns = set()
+    if respect_gitignore:
+        gitignore_patterns = _load_gitignore_patterns(search_path)
+    
+    # Get files to search
+    if os.path.isfile(search_path):
+        files_to_search = [search_path]
+    else:
+        files_to_search = []
+        for root, dirs, files in os.walk(search_path):
+            # Filter directories in-place to skip ignored ones
+            if respect_gitignore:
+                dirs[:] = [d for d in dirs if not _should_ignore(
+                    os.path.join(root, d), search_path, gitignore_patterns
+                )]
+            
+            for filename in files:
+                filepath = os.path.join(root, filename)
+                
+                # Check gitignore
+                if respect_gitignore and _should_ignore(filepath, search_path, gitignore_patterns):
+                    continue
+                
+                # Check include patterns
+                if include_patterns:
+                    rel_path = os.path.relpath(filepath, search_path)
+                    if not any(fnmatch.fnmatch(rel_path, p) or fnmatch.fnmatch(filename, p) 
+                              for p in include_patterns):
+                        continue
+                
+                # Check exclude patterns
+                if exclude_patterns:
+                    rel_path = os.path.relpath(filepath, search_path)
+                    if any(fnmatch.fnmatch(rel_path, p) or fnmatch.fnmatch(filename, p) 
+                          for p in exclude_patterns):
+                        continue
+                
+                files_to_search.append(filepath)
+    
+    # Search files
+    for filepath in files_to_search:
+        if len(results) >= max_results:
+            break
+        
+        # Skip binary files
+        if _is_binary_file(filepath):
+            continue
+        
+        try:
+            with open(filepath, 'r', encoding='utf-8', errors='ignore') as f:
+                lines = f.readlines()
+            
+            for i, line in enumerate(lines):
+                if len(results) >= max_results:
+                    break
+                
+                if regex.search(line):
+                    # Get context
+                    start_ctx = max(0, i - context_lines)
+                    end_ctx = min(len(lines), i + context_lines + 1)
+                    context = ''.join(lines[start_ctx:end_ctx])
+                    
+                    results.append({
+                        "path": os.path.relpath(filepath, search_path),
+                        "absolute_path": filepath,
+                        "line_number": i + 1,  # 1-indexed
+                        "content": line.rstrip('\n\r'),
+                        "context": context.rstrip('\n\r') if context_lines > 0 else None,
+                        "context_start": start_ctx + 1 if context_lines > 0 else None,
+                        "context_end": end_ctx if context_lines > 0 else None
+                    })
+        except Exception as e:
+            logger.debug(f"Error reading {filepath}: {e}")
+            continue
+    
+    return results
+
+
+def glob_search(
+    search_path: str,
+    pattern: str,
+    max_results: int = 100,
+    include_dirs: bool = False,
+    respect_gitignore: bool = True
+) -> List[Dict[str, Any]]:
+    """Search for files matching glob pattern.
+    
+    Args:
+        search_path: Directory to search
+        pattern: Glob pattern (e.g., "**/*.py", "src/*.js")
+        max_results: Maximum number of results
+        include_dirs: If True, include directories in results
+        respect_gitignore: If True, respect .gitignore patterns
+        
+    Returns:
+        List of file info dictionaries with path, size, is_dir
+    """
+    results = []
+    search_path = os.path.abspath(search_path)
+    
+    # Load gitignore patterns
+    gitignore_patterns = set()
+    if respect_gitignore:
+        gitignore_patterns = _load_gitignore_patterns(search_path)
+    
+    # Use pathlib for glob
+    root = Path(search_path)
+    
+    try:
+        for match in root.glob(pattern):
+            if len(results) >= max_results:
+                break
+            
+            # Check gitignore
+            if respect_gitignore and _should_ignore(str(match), search_path, gitignore_patterns):
+                continue
+            
+            # Skip directories unless requested
+            if match.is_dir() and not include_dirs:
+                continue
+            
+            # Skip binary files for file results
+            if match.is_file() and _is_binary_file(str(match)):
+                continue
+            
+            try:
+                stat = match.stat()
+                results.append({
+                    "path": str(match.relative_to(root)),
+                    "absolute_path": str(match),
+                    "size": stat.st_size if match.is_file() else None,
+                    "is_dir": match.is_dir(),
+                    "modified": stat.st_mtime
+                })
+            except Exception:
+                continue
+    except Exception as e:
+        logger.debug(f"Error in glob search: {e}")
+    
+    return results
+
+
+def read_file(
+    filepath: str,
+    start_line: Optional[int] = None,
+    end_line: Optional[int] = None,
+    context_lines: int = 0,
+    max_lines: int = 500
+) -> Dict[str, Any]:
+    """Read file contents with optional line range.
+    
+    Args:
+        filepath: Path to the file
+        start_line: Starting line (1-indexed, inclusive)
+        end_line: Ending line (1-indexed, inclusive)
+        context_lines: Additional context lines before/after range
+        max_lines: Maximum lines to return
+        
+    Returns:
+        Dictionary with content, line info, and metadata
+    """
+    filepath = os.path.abspath(filepath)
+    
+    if not os.path.exists(filepath):
+        return {
+            "success": False,
+            "error": f"File not found: {filepath}",
+            "path": filepath
+        }
+    
+    if os.path.isdir(filepath):
+        return {
+            "success": False,
+            "error": f"Path is a directory: {filepath}",
+            "path": filepath
+        }
+    
+    if _is_binary_file(filepath):
+        return {
+            "success": False,
+            "error": f"Binary file: {filepath}",
+            "path": filepath
+        }
+    
+    try:
+        with open(filepath, 'r', encoding='utf-8', errors='ignore') as f:
+            lines = f.readlines()
+        
+        total_lines = len(lines)
+        
+        # Determine line range
+        if start_line is None:
+            start_line = 1
+        if end_line is None:
+            end_line = total_lines
+        
+        # Apply context
+        actual_start = max(1, start_line - context_lines)
+        actual_end = min(total_lines, end_line + context_lines)
+        
+        # Limit lines
+        if actual_end - actual_start + 1 > max_lines:
+            actual_end = actual_start + max_lines - 1
+        
+        # Extract content (convert to 0-indexed)
+        content_lines = lines[actual_start - 1:actual_end]
+        content = ''.join(content_lines)
+        
+        return {
+            "success": True,
+            "path": filepath,
+            "content": content.rstrip('\n'),
+            "start_line": actual_start,
+            "end_line": actual_end,
+            "total_lines": total_lines,
+            "lines_returned": len(content_lines)
+        }
+    except Exception as e:
+        return {
+            "success": False,
+            "error": str(e),
+            "path": filepath
+        }
+
+
+def list_directory(
+    dir_path: str,
+    recursive: bool = False,
+    max_depth: int = 10,
+    max_entries: int = 500,
+    include_hidden: bool = False,
+    respect_gitignore: bool = True
+) -> Dict[str, Any]:
+    """List directory contents.
+    
+    Args:
+        dir_path: Directory path
+        recursive: If True, list recursively
+        max_depth: Maximum recursion depth
+        max_entries: Maximum entries to return
+        include_hidden: If True, include hidden files/dirs
+        respect_gitignore: If True, respect .gitignore patterns
+        
+    Returns:
+        Dictionary with entries list and metadata
+    """
+    dir_path = os.path.abspath(dir_path)
+    
+    if not os.path.exists(dir_path):
+        return {
+            "success": False,
+            "error": f"Directory not found: {dir_path}",
+            "path": dir_path
+        }
+    
+    if not os.path.isdir(dir_path):
+        return {
+            "success": False,
+            "error": f"Not a directory: {dir_path}",
+            "path": dir_path
+        }
+    
+    # Load gitignore patterns
+    gitignore_patterns = set()
+    if respect_gitignore:
+        gitignore_patterns = _load_gitignore_patterns(dir_path)
+    
+    entries = []
+    
+    def _list_dir(path: str, depth: int):
+        if depth > max_depth or len(entries) >= max_entries:
+            return
+        
+        try:
+            for entry in os.scandir(path):
+                if len(entries) >= max_entries:
+                    break
+                
+                # Skip hidden files unless requested
+                if not include_hidden and entry.name.startswith('.'):
+                    continue
+                
+                # Check gitignore
+                if respect_gitignore and _should_ignore(entry.path, dir_path, gitignore_patterns):
+                    continue
+                
+                try:
+                    stat = entry.stat()
+                    entries.append({
+                        "name": entry.name,
+                        "path": os.path.relpath(entry.path, dir_path),
+                        "absolute_path": entry.path,
+                        "is_dir": entry.is_dir(),
+                        "size": stat.st_size if entry.is_file() else None,
+                        "modified": stat.st_mtime
+                    })
+                except Exception:
+                    continue
+                
+                # Recurse into directories
+                if recursive and entry.is_dir():
+                    _list_dir(entry.path, depth + 1)
+        except PermissionError:
+            pass
+        except Exception as e:
+            logger.debug(f"Error listing {path}: {e}")
+    
+    _list_dir(dir_path, 0)
+    
+    # Sort entries: directories first, then by name
+    entries.sort(key=lambda e: (not e["is_dir"], e["name"].lower()))
+    
+    return {
+        "success": True,
+        "path": dir_path,
+        "entries": entries,
+        "total_entries": len(entries),
+        "truncated": len(entries) >= max_entries
+    }
+
+
+# Tool definitions for LLM function calling
+FAST_CONTEXT_TOOLS = [
+    {
+        "name": "grep_search",
+        "description": "Search for a pattern in files within a directory. Returns matching lines with file paths and line numbers.",
+        "parameters": {
+            "type": "object",
+            "properties": {
+                "search_path": {
+                    "type": "string",
+                    "description": "Directory or file path to search"
+                },
+                "pattern": {
+                    "type": "string",
+                    "description": "Search pattern (text or regex)"
+                },
+                "is_regex": {
+                    "type": "boolean",
+                    "description": "If true, treat pattern as regex",
+                    "default": False
+                },
+                "case_sensitive": {
+                    "type": "boolean",
+                    "description": "If true, search is case sensitive",
+                    "default": False
+                },
+                "include_patterns": {
+                    "type": "array",
+                    "items": {"type": "string"},
+                    "description": "Glob patterns to include (e.g., ['*.py'])"
+                },
+                "max_results": {
+                    "type": "integer",
+                    "description": "Maximum results to return",
+                    "default": 50
+                }
+            },
+            "required": ["search_path", "pattern"]
+        }
+    },
+    {
+        "name": "glob_search",
+        "description": "Find files matching a glob pattern. Returns file paths and metadata.",
+        "parameters": {
+            "type": "object",
+            "properties": {
+                "search_path": {
+                    "type": "string",
+                    "description": "Directory to search"
+                },
+                "pattern": {
+                    "type": "string",
+                    "description": "Glob pattern (e.g., '**/*.py', 'src/*.js')"
+                },
+                "max_results": {
+                    "type": "integer",
+                    "description": "Maximum results to return",
+                    "default": 50
+                }
+            },
+            "required": ["search_path", "pattern"]
+        }
+    },
+    {
+        "name": "read_file",
+        "description": "Read contents of a file, optionally with specific line range.",
+        "parameters": {
+            "type": "object",
+            "properties": {
+                "filepath": {
+                    "type": "string",
+                    "description": "Path to the file"
+                },
+                "start_line": {
+                    "type": "integer",
+                    "description": "Starting line number (1-indexed)"
+                },
+                "end_line": {
+                    "type": "integer",
+                    "description": "Ending line number (1-indexed)"
+                },
+                "context_lines": {
+                    "type": "integer",
+                    "description": "Additional context lines",
+                    "default": 0
+                }
+            },
+            "required": ["filepath"]
+        }
+    },
+    {
+        "name": "list_directory",
+        "description": "List contents of a directory.",
+        "parameters": {
+            "type": "object",
+            "properties": {
+                "dir_path": {
+                    "type": "string",
+                    "description": "Directory path"
+                },
+                "recursive": {
+                    "type": "boolean",
+                    "description": "If true, list recursively",
+                    "default": False
+                },
+                "max_depth": {
+                    "type": "integer",
+                    "description": "Maximum recursion depth",
+                    "default": 3
+                }
+            },
+            "required": ["dir_path"]
+        }
+    }
+]
+
+
+def execute_tool(tool_name: str, **kwargs) -> Dict[str, Any]:
+    """Execute a search tool by name.
+    
+    Args:
+        tool_name: Name of the tool to execute
+        **kwargs: Tool arguments
+        
+    Returns:
+        Tool result dictionary
+    """
+    tools = {
+        "grep_search": grep_search,
+        "glob_search": glob_search,
+        "read_file": read_file,
+        "list_directory": list_directory
+    }
+    
+    if tool_name not in tools:
+        return {"error": f"Unknown tool: {tool_name}"}
+    
+    try:
+        return tools[tool_name](**kwargs)
+    except Exception as e:
+        return {"error": str(e)}

src/praisonai-agents/praisonaiagents/llm/llm.py

@@ -239,6 +239,10 @@ class LLM:
         api_key: Optional[str] = None,
         base_url: Optional[str] = None,
         events: List[Any] = [],
+        web_search: Optional[Union[bool, Dict[str, Any]]] = None,
+        web_fetch: Optional[Union[bool, Dict[str, Any]]] = None,
+        prompt_caching: Optional[bool] = None,
+        claude_memory: Optional[Union[bool, Any]] = None,
         **extra_settings
     ):
         # Configure logging only once at the class level
@@ -273,6 +277,11 @@ class LLM:
         self.base_url = base_url
         self.events = events
         self.extra_settings = extra_settings
+        self.web_search = web_search
+        self.web_fetch = web_fetch
+        self.prompt_caching = prompt_caching
+        self.claude_memory = claude_memory
+        self._claude_memory_tool = None  # Lazy initialized
         self._console = None  # Lazy load console when needed
         self.chat_history = []
         self.verbose = extra_settings.get('verbose', True)
@@ -381,6 +390,109 @@ class LLM:
         else:
             return False
 
+    def _supports_web_search(self) -> bool:
+        """
+        Check if the current model supports native web search via LiteLLM.
+        
+        Native web search allows the model to search the web in real-time
+        without requiring external tools like DuckDuckGo.
+        
+        Returns:
+            bool: True if the model supports native web search, False otherwise
+        """
+        from .model_capabilities import supports_web_search
+        return supports_web_search(self.model)
+
+    def _supports_web_fetch(self) -> bool:
+        """
+        Check if the current model supports web fetch via LiteLLM.
+        
+        Web fetch allows the model to retrieve full content from specific URLs
+        (web pages and PDF documents). Currently only supported by Anthropic Claude models.
+        
+        Returns:
+            bool: True if the model supports web fetch, False otherwise
+        """
+        from .model_capabilities import supports_web_fetch
+        return supports_web_fetch(self.model)
+
+    def _supports_prompt_caching(self) -> bool:
+        """
+        Check if the current model supports prompt caching via LiteLLM.
+        
+        Prompt caching allows caching parts of prompts to reduce costs and latency.
+        Supported by OpenAI, Anthropic, Bedrock, and Deepseek.
+        
+        Returns:
+            bool: True if the model supports prompt caching, False otherwise
+        """
+        from .model_capabilities import supports_prompt_caching
+        return supports_prompt_caching(self.model)
+
+    def _get_claude_memory_tool(self):
+        """
+        Get or initialize the Claude Memory Tool instance.
+        
+        Returns:
+            ClaudeMemoryTool instance if claude_memory is enabled, None otherwise
+        """
+        if not self.claude_memory:
+            return None
+        
+        if self._claude_memory_tool is None:
+            from ..tools.claude_memory_tool import ClaudeMemoryTool
+            
+            if isinstance(self.claude_memory, bool):
+                # Create default memory tool
+                self._claude_memory_tool = ClaudeMemoryTool()
+            else:
+                # Use provided memory tool instance
+                self._claude_memory_tool = self.claude_memory
+        
+        return self._claude_memory_tool
+
+    def _supports_claude_memory(self) -> bool:
+        """
+        Check if the current model supports Claude Memory Tool.
+        
+        Claude Memory Tool is only supported by Anthropic Claude models.
+        
+        Returns:
+            bool: True if the model supports Claude Memory Tool
+        """
+        if not self.claude_memory:
+            return False
+        
+        return self._is_anthropic_model()
+
+    def _is_memory_tool_call(self, function_name: str) -> bool:
+        """
+        Check if a tool call is for the Claude Memory Tool.
+        
+        Args:
+            function_name: Name of the function being called
+            
+        Returns:
+            bool: True if this is a memory tool call
+        """
+        return function_name == "memory" and self._supports_claude_memory()
+
+    def _execute_memory_tool_call(self, arguments: Dict[str, Any]) -> str:
+        """
+        Execute a Claude Memory Tool call.
+        
+        Args:
+            arguments: The arguments from Claude's tool call
+            
+        Returns:
+            Result string from the memory operation
+        """
+        memory_tool = self._get_claude_memory_tool()
+        if not memory_tool:
+            return "Error: Memory tool not initialized"
+        
+        return memory_tool.process_tool_call(arguments)
+
     def _generate_ollama_tool_summary(self, tool_results: List[Any], response_text: str) -> Optional[str]:
         """
         Generate a summary from tool results for Ollama to prevent infinite loops.
@@ -724,7 +836,21 @@ class LLM:
             
             # Skip system messages for legacy o1 models as they don't support them
             if not self._needs_system_message_skip():
-                messages.append({"role": "system", "content": system_prompt})
+                # Apply prompt caching for Anthropic models if enabled
+                if self.prompt_caching and self._supports_prompt_caching() and self._is_anthropic_model():
+                    # Anthropic requires cache_control in content array format
+                    messages.append({
+                        "role": "system",
+                        "content": [
+                            {
+                                "type": "text",
+                                "text": system_prompt,
+                                "cache_control": {"type": "ephemeral"}
+                            }
+                        ]
+                    })
+                else:
+                    messages.append({"role": "system", "content": system_prompt})
         
         # Add chat history if provided
         if chat_history:
@@ -2295,6 +2421,12 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
             return False
         return any(prefix in self.model.lower() for prefix in ['gemini', 'gemini/', 'google/gemini'])
     
+    def _is_anthropic_model(self) -> bool:
+        """Check if the model is an Anthropic Claude model."""
+        if not self.model:
+            return False
+        return any(prefix in self.model.lower() for prefix in ['claude', 'anthropic/', 'anthropic/claude'])
+    
     def _is_streaming_error_recoverable(self, error: Exception) -> bool:
         """Check if a streaming error is recoverable (e.g., malformed chunk vs connection error)."""
         error_msg = str(error).lower()
@@ -3171,6 +3303,79 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                     logging.debug(f"Could not verify function calling support: {e}. Setting tool_choice anyway.")
                     params['tool_choice'] = 'auto'
         
+        # Add web_search_options if web_search is enabled and model supports it
+        if self.web_search and self._supports_web_search():
+            if isinstance(self.web_search, dict):
+                # Use custom web_search_options provided by user
+                params['web_search_options'] = self.web_search
+            else:
+                # Use default web_search_options
+                params['web_search_options'] = {"search_context_size": "medium"}
+            logging.debug(f"Web search enabled with options: {params['web_search_options']}")
+            
+            # Some web search models (like gpt-4o-search-preview) don't support certain parameters
+            # Remove incompatible parameters to avoid BadRequestError
+            incompatible_params = ['temperature', 'top_p', 'presence_penalty', 'frequency_penalty', 'logit_bias']
+            for param in incompatible_params:
+                if param in params:
+                    del params[param]
+                    logging.debug(f"Removed incompatible parameter '{param}' for web search model")
+        
+        # Add web_fetch tool if web_fetch is enabled and model supports it
+        # Web fetch is passed as a tool (unlike web_search which uses web_search_options)
+        if self.web_fetch and self._supports_web_fetch():
+            # Build the web_fetch tool definition
+            web_fetch_tool = {
+                "type": "web_fetch_20250910",
+                "name": "web_fetch",
+            }
+            
+            # Add optional parameters from web_fetch dict if provided
+            if isinstance(self.web_fetch, dict):
+                if 'max_uses' in self.web_fetch:
+                    web_fetch_tool['max_uses'] = self.web_fetch['max_uses']
+                if 'allowed_domains' in self.web_fetch:
+                    web_fetch_tool['allowed_domains'] = self.web_fetch['allowed_domains']
+                if 'blocked_domains' in self.web_fetch:
+                    web_fetch_tool['blocked_domains'] = self.web_fetch['blocked_domains']
+                if 'citations' in self.web_fetch:
+                    web_fetch_tool['citations'] = self.web_fetch['citations']
+                if 'max_content_tokens' in self.web_fetch:
+                    web_fetch_tool['max_content_tokens'] = self.web_fetch['max_content_tokens']
+            else:
+                # Default max_uses if web_fetch=True
+                web_fetch_tool['max_uses'] = 5
+            
+            # Add web_fetch tool to existing tools or create tools list
+            if 'tools' in params and params['tools']:
+                params['tools'].append(web_fetch_tool)
+            else:
+                params['tools'] = [web_fetch_tool]
+            
+            logging.debug(f"Web fetch enabled with tool: {web_fetch_tool}")
+        
+        # Add Claude Memory Tool if enabled and model supports it
+        if self._supports_claude_memory():
+            memory_tool = self._get_claude_memory_tool()
+            if memory_tool:
+                # Add the memory tool definition
+                memory_tool_def = memory_tool.get_tool_definition()
+                
+                # Add memory tool to existing tools or create tools list
+                if 'tools' in params and params['tools']:
+                    params['tools'].append(memory_tool_def)
+                else:
+                    params['tools'] = [memory_tool_def]
+                
+                # Add the beta header for Anthropic
+                beta_header = memory_tool.get_beta_header()
+                if 'extra_headers' in params:
+                    params['extra_headers']['anthropic-beta'] = beta_header
+                else:
+                    params['extra_headers'] = {'anthropic-beta': beta_header}
+                
+                logging.debug(f"Claude memory tool enabled with beta header: {beta_header}")
+        
         return params
 
     def _prepare_response_logging(self, temperature: float, stream: bool, verbose: bool, markdown: bool, **kwargs) -> Optional[Dict[str, Any]]:

src/praisonai-agents/praisonaiagents/llm/model_capabilities.py

@@ -1,62 +1,30 @@
 """
 Model capabilities configuration for different LLM providers.
-This module defines which models support specific features like structured outputs.
+
+This module uses LiteLLM's helper functions as the primary source for model capability detection.
+LiteLLM is maintained by many contributors and is more accurate and up-to-date.
+
+LiteLLM Helper Functions:
+- litellm.supports_web_search(model=) - Check web search support
+- litellm.supports_function_calling(model=) - Check function calling support
+- litellm.supports_parallel_function_calling(model=) - Check parallel function calling
+- litellm.supports_response_schema(model=) - Check JSON schema/structured outputs support
+- litellm.utils.supports_prompt_caching(model=) - Check prompt caching support
+- litellm.get_supported_openai_params(model=) - Get all supported params
+
+Sources:
+- https://docs.litellm.ai/docs/completion/web_search
+- https://docs.litellm.ai/docs/completion/json_mode
+- https://docs.litellm.ai/docs/completion/function_call
+- https://docs.litellm.ai/docs/completion/prompt_caching
 """
 
-# Models that support OpenAI-style structured outputs (response_format with Pydantic models)
-MODELS_SUPPORTING_STRUCTURED_OUTPUTS = {
-    # OpenAI models
-    "gpt-5-nano",
-    "gpt-5-nano",
-    "gpt-4-turbo",
-    "gpt-4-turbo-preview",
-    "gpt-4-turbo-2024-04-09",
-    "gpt-4-1106-preview",
-    "gpt-4-0125-preview",
-    "gpt-3.5-turbo",
-    "gpt-3.5-turbo-1106",
-    "gpt-3.5-turbo-0125",
-    
-    # New/Future OpenAI models (as mentioned by user)
-    "codex-mini",
-    "o3-pro",
-    "gpt-4.5-preview",
-    "o3-mini",
-    "o1",
-    "o1-preview",
-    "o1-mini",
-    "gpt-4.1",
-    "gpt-4.1-nano",
-    "gpt-4.1-mini",
-    "o4-mini",
-    "o3",
-    
-    # Gemini models that support structured outputs
-    "gemini-2.0-flash",
-    "gemini-2.0-flash-exp",
-    "gemini-1.5-pro",
-    "gemini-1.5-pro-latest",
-    "gemini-1.5-flash",
-    "gemini-1.5-flash-latest",
-    "gemini-1.5-flash-8b",
-    "gemini-1.5-flash-8b-latest",
-}
-
-# Models that explicitly DON'T support structured outputs
-MODELS_NOT_SUPPORTING_STRUCTURED_OUTPUTS = {
-    # Audio preview models
-    "gpt-4o-audio-preview",
-    "gpt-5-nano-audio-preview",
-    
-    # Legacy o1 models (don't support system messages either)
-    "o1-preview-2024-09-12",
-    "o1-mini-2024-09-12",
-}
-
 
 def supports_structured_outputs(model_name: str) -> bool:
     """
-    Check if a model supports OpenAI-style structured outputs.
+    Check if a model supports structured outputs (JSON schema).
+    
+    Uses LiteLLM's supports_response_schema() as the primary check.
     
     Args:
         model_name: The name of the model to check
@@ -67,34 +35,72 @@ def supports_structured_outputs(model_name: str) -> bool:
     if not model_name:
         return False
     
-    # Strip provider prefixes (e.g., 'google/', 'openai/', etc.)
-    model_without_provider = model_name
-    for prefix in ['google/', 'openai/', 'anthropic/', 'gemini/', 'mistral/', 'deepseek/', 'groq/']:
-        if model_name.startswith(prefix):
-            model_without_provider = model_name[len(prefix):]
-            break
+    try:
+        import litellm
+        # Use LiteLLM's built-in check - most accurate and up-to-date
+        if hasattr(litellm, 'supports_response_schema'):
+            return litellm.supports_response_schema(model=model_name)
+    except Exception:
+        pass
     
-    # First check if it's explicitly in the NOT supporting list
-    if model_without_provider in MODELS_NOT_SUPPORTING_STRUCTURED_OUTPUTS:
+    return False
+
+
+def supports_function_calling(model_name: str) -> bool:
+    """
+    Check if a model supports function calling.
+    
+    Uses LiteLLM's supports_function_calling() as the primary check.
+    
+    Args:
+        model_name: The name of the model to check
+        
+    Returns:
+        bool: True if the model supports function calling, False otherwise
+    """
+    if not model_name:
         return False
     
-    # Then check if it's in the supporting list
-    if model_without_provider in MODELS_SUPPORTING_STRUCTURED_OUTPUTS:
-        return True
+    try:
+        import litellm
+        # Use LiteLLM's built-in check - most accurate and up-to-date
+        if hasattr(litellm, 'supports_function_calling'):
+            return litellm.supports_function_calling(model=model_name)
+    except Exception:
+        pass
     
-    # For models with version suffixes, check the base model name
-    base_model = model_without_provider.split('-2024-')[0].split('-2025-')[0]
-    if base_model in MODELS_SUPPORTING_STRUCTURED_OUTPUTS:
-        return True
+    return False
+
+
+def supports_parallel_function_calling(model_name: str) -> bool:
+    """
+    Check if a model supports parallel function calling.
+    
+    Uses LiteLLM's supports_parallel_function_calling() as the primary check.
+    
+    Args:
+        model_name: The name of the model to check
+        
+    Returns:
+        bool: True if the model supports parallel function calling, False otherwise
+    """
+    if not model_name:
+        return False
+    
+    try:
+        import litellm
+        # Use LiteLLM's built-in check - most accurate and up-to-date
+        if hasattr(litellm, 'supports_parallel_function_calling'):
+            return litellm.supports_parallel_function_calling(model=model_name)
+    except Exception:
+        pass
     
-    # Default to False for unknown models
     return False
 
 
 def supports_streaming_with_tools(model_name: str) -> bool:
     """
     Check if a model supports streaming when tools are provided.
-    Most models that support structured outputs also support streaming with tools.
     
     Args:
         model_name: The name of the model to check
@@ -102,15 +108,153 @@ def supports_streaming_with_tools(model_name: str) -> bool:
     Returns:
         bool: True if the model supports streaming with tools, False otherwise
     """
-    # For now, use the same logic as structured outputs
-    # In the future, this could be a separate list if needed
-    return supports_structured_outputs(model_name)
+    # Models that support function calling generally support streaming with tools
+    return supports_function_calling(model_name)
 
 
 # Supported Gemini internal tools
 GEMINI_INTERNAL_TOOLS = {'googleSearch', 'urlContext', 'codeExecution'}
 
 
+def supports_web_search(model_name: str) -> bool:
+    """
+    Check if a model supports native web search via LiteLLM.
+    
+    Uses LiteLLM's supports_web_search() as the primary check.
+    
+    Native web search allows the model to search the web in real-time
+    without requiring external tools like DuckDuckGo.
+    
+    Supported providers:
+    - OpenAI (gpt-4o-search-preview, gpt-4o-mini-search-preview)
+    - xAI (grok-3)
+    - Anthropic (claude-3-5-sonnet-latest, claude-sonnet-4, etc.)
+    - Google/Vertex AI (gemini-2.0-flash, gemini-2.5-*, etc.)
+    - Perplexity (all models)
+    
+    Args:
+        model_name: The name of the model to check (with or without provider prefix)
+        
+    Returns:
+        bool: True if the model supports native web search, False otherwise
+    """
+    if not model_name:
+        return False
+    
+    try:
+        import litellm
+        # Use LiteLLM's built-in check - most accurate and up-to-date
+        if hasattr(litellm, 'supports_web_search'):
+            return litellm.supports_web_search(model=model_name)
+    except Exception:
+        pass
+    
+    return False
+
+
+def supports_prompt_caching(model_name: str) -> bool:
+    """
+    Check if a model supports prompt caching via LiteLLM.
+    
+    Uses LiteLLM's supports_prompt_caching() as the primary check.
+    
+    Prompt caching allows caching parts of prompts to reduce costs and latency
+    on subsequent requests with similar prompts.
+    
+    Supported providers:
+    - OpenAI (openai/) - Automatic caching for prompts ≥1024 tokens
+    - Anthropic (anthropic/) - Manual caching with cache_control
+    - Bedrock (bedrock/) - All models that support prompt caching
+    - Deepseek (deepseek/) - Works like OpenAI
+    
+    Args:
+        model_name: The name of the model to check (with or without provider prefix)
+        
+    Returns:
+        bool: True if the model supports prompt caching, False otherwise
+    """
+    if not model_name:
+        return False
+    
+    try:
+        from litellm.utils import supports_prompt_caching as litellm_supports_prompt_caching
+        return litellm_supports_prompt_caching(model=model_name)
+    except Exception:
+        pass
+    
+    return False
+
+
+# Models that support web fetch via LiteLLM (Anthropic only)
+# Web fetch retrieves full content from specific URLs (web pages and PDFs)
+# Source: https://docs.litellm.ai/docs/completion/web_fetch
+# Note: LiteLLM doesn't have a supports_web_fetch() helper yet, so we maintain a static list
+MODELS_SUPPORTING_WEB_FETCH = {
+    # Anthropic Claude models with web fetch support
+    "claude-opus-4-1-20250805",
+    "claude-opus-4-1",
+    "claude-opus-4-20250514",
+    "claude-opus-4",
+    "claude-sonnet-4-20250514",
+    "claude-sonnet-4",
+    "claude-3-7-sonnet-20250219",
+    "claude-3-7-sonnet-latest",
+    "claude-3-5-sonnet-latest",
+    "claude-3-5-sonnet-20241022",
+    "claude-3-5-haiku-latest",
+    "claude-3-5-haiku-20241022",
+}
+
+
+def supports_web_fetch(model_name: str) -> bool:
+    """
+    Check if a model supports web fetch via LiteLLM.
+    
+    Web fetch allows the model to retrieve full content from specific URLs
+    (web pages and PDF documents). Currently only supported by Anthropic Claude models.
+    
+    Note: LiteLLM doesn't have a supports_web_fetch() helper yet, so we use a static list
+    with auto-detection for Claude 4+ models.
+    
+    Args:
+        model_name: The name of the model to check (with or without provider prefix)
+        
+    Returns:
+        bool: True if the model supports web fetch, False otherwise
+    """
+    if not model_name:
+        return False
+    
+    # Strip provider prefixes
+    model_without_provider = model_name
+    for prefix in ['anthropic/', 'bedrock/', 'vertex_ai/']:
+        if model_name.startswith(prefix):
+            model_without_provider = model_name[len(prefix):]
+            break
+    
+    # Check our static list
+    if model_without_provider in MODELS_SUPPORTING_WEB_FETCH:
+        return True
+    
+    # Check base model name (without version suffix)
+    base_model = model_without_provider.split('-2024-')[0].split('-2025-')[0]
+    if base_model in MODELS_SUPPORTING_WEB_FETCH:
+        return True
+    
+    # Auto-support for Claude 4+ models (Anthropic only)
+    model_lower = model_without_provider.lower()
+    if 'claude' in model_lower:
+        import re
+        # Match patterns like claude-4, claude-5, claude-sonnet-4, claude-opus-4, etc.
+        version_match = re.search(r'claude-(?:sonnet-|opus-|haiku-)?(\d+)', model_lower)
+        if version_match:
+            version = int(version_match.group(1))
+            if version >= 4:  # Claude 4 and later
+                return True
+    
+    return False
+
+
 def is_gemini_internal_tool(tool) -> bool:
     """
     Check if a tool is a Gemini internal tool and should be included in formatted tools.

src/praisonai-agents/praisonaiagents/mcp/__init__.py

@@ -2,8 +2,65 @@
 Model Context Protocol (MCP) integration for PraisonAI Agents.
 
 This package provides classes and utilities for connecting to MCP servers
-using different transport methods (stdio, SSE, etc.).
+using different transport methods:
+- stdio: Standard input/output (subprocess)
+- sse: Server-Sent Events (legacy HTTP+SSE)
+- http_stream: Streamable HTTP (current standard)
+- websocket: WebSocket (SEP-1288)
+
+Protocol Revision: 2025-11-25
 """
 from .mcp import MCP
 
-__all__ = ["MCP"]
+# Lazy imports for optional components
+def __getattr__(name):
+    """Lazy import for optional modules."""
+    if name == "WebSocketMCPClient":
+        from .mcp_websocket import WebSocketMCPClient
+        return WebSocketMCPClient
+    elif name == "SessionManager":
+        from .mcp_session import SessionManager
+        return SessionManager
+    elif name == "SecurityConfig":
+        from .mcp_security import SecurityConfig
+        return SecurityConfig
+    elif name == "BaseTransport":
+        from .mcp_transport import BaseTransport
+        return BaseTransport
+    elif name == "TransportConfig":
+        from .mcp_transport import TransportConfig
+        return TransportConfig
+    elif name == "ToolsMCPServer":
+        from .mcp_server import ToolsMCPServer
+        return ToolsMCPServer
+    elif name == "launch_tools_mcp_server":
+        from .mcp_server import launch_tools_mcp_server
+        return launch_tools_mcp_server
+    elif name == "function_to_mcp_schema":
+        from .mcp_utils import function_to_mcp_schema
+        return function_to_mcp_schema
+    elif name == "get_tool_metadata":
+        from .mcp_utils import get_tool_metadata
+        return get_tool_metadata
+    elif name == "python_type_to_json_schema":
+        from .mcp_utils import python_type_to_json_schema
+        return python_type_to_json_schema
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+__all__ = [
+    # Client
+    "MCP",
+    # Server (lazy loaded)
+    "ToolsMCPServer",
+    "launch_tools_mcp_server",
+    # Utilities (lazy loaded)
+    "function_to_mcp_schema",
+    "get_tool_metadata",
+    "python_type_to_json_schema",
+    # Transport components (lazy loaded)
+    "WebSocketMCPClient",
+    "SessionManager",
+    "SecurityConfig",
+    "BaseTransport",
+    "TransportConfig",
+]

src/praisonai-agents/praisonaiagents/mcp/mcp.py

@@ -227,6 +227,26 @@ class MCP:
         self.timeout = timeout
         self.debug = debug
         
+        # Check if this is a WebSocket URL (ws:// or wss://)
+        if isinstance(command_or_string, str) and re.match(r'^wss?://', command_or_string):
+            from .mcp_websocket import WebSocketMCPClient
+            # Extract auth token if provided
+            auth_token = kwargs.pop('auth_token', None)
+            
+            self.websocket_client = WebSocketMCPClient(
+                command_or_string,
+                debug=debug,
+                timeout=timeout,
+                auth_token=auth_token,
+                options=kwargs
+            )
+            self._tools = list(self.websocket_client.tools)
+            self.is_sse = False
+            self.is_http_stream = False
+            self.is_websocket = True
+            self.is_npx = False
+            return
+        
         # Check if this is an HTTP URL
         if isinstance(command_or_string, str) and re.match(r'^https?://', command_or_string):
             # Determine transport type based on URL or kwargs

src/praisonai-agents/praisonaiagents/mcp/mcp_compat.py

@@ -0,0 +1,150 @@
+"""
+MCP Backward Compatibility Module.
+
+This module provides utilities for maintaining backward compatibility
+with older MCP servers and the deprecated HTTP+SSE transport from
+protocol version 2024-11-05.
+
+Per MCP specification, clients wanting to support older servers should:
+1. Attempt POST to server URL with InitializeRequest
+2. If it fails with 400, 404, or 405, fall back to legacy SSE
+3. Issue GET to server URL expecting SSE stream with endpoint event
+"""
+
+from typing import Optional
+
+
+def detect_transport_support(
+    response_status: int,
+    content_type: Optional[str] = None
+) -> str:
+    """
+    Detect which transport a server supports based on response.
+    
+    Per MCP spec backward compatibility guide:
+    - Success (200) with JSON indicates Streamable HTTP
+    - 400, 404, 405 indicates need for legacy SSE fallback
+    
+    Args:
+        response_status: HTTP response status code
+        content_type: Response Content-Type header
+        
+    Returns:
+        Transport type: "http_stream" or "sse"
+    """
+    # Check for fallback indicators
+    if response_status in (400, 404, 405):
+        return "sse"
+    
+    # Successful response indicates Streamable HTTP support
+    if response_status == 200:
+        return "http_stream"
+    
+    # Default to SSE for unknown cases
+    return "sse"
+
+
+def is_legacy_sse_url(url: str) -> bool:
+    """
+    Check if URL indicates legacy SSE transport.
+    
+    Legacy SSE URLs typically end with /sse.
+    
+    Args:
+        url: Server URL
+        
+    Returns:
+        True if URL indicates legacy SSE
+    """
+    return url.rstrip('/').endswith('/sse')
+
+
+def get_legacy_sse_endpoint(base_url: str) -> str:
+    """
+    Get the legacy SSE endpoint URL.
+    
+    Args:
+        base_url: Base server URL
+        
+    Returns:
+        SSE endpoint URL
+    """
+    base = base_url.rstrip('/')
+    if not base.endswith('/sse'):
+        return f"{base}/sse"
+    return base
+
+
+def get_streamable_http_endpoint(base_url: str) -> str:
+    """
+    Get the Streamable HTTP endpoint URL.
+    
+    Args:
+        base_url: Base server URL
+        
+    Returns:
+        MCP endpoint URL
+    """
+    base = base_url.rstrip('/')
+    # Remove /sse suffix if present
+    if base.endswith('/sse'):
+        base = base[:-4]
+    if not base.endswith('/mcp'):
+        return f"{base}/mcp"
+    return base
+
+
+class TransportNegotiator:
+    """
+    Handles automatic transport negotiation with MCP servers.
+    
+    This class implements the backward compatibility algorithm
+    from the MCP specification for detecting server capabilities.
+    """
+    
+    def __init__(self, base_url: str):
+        """
+        Initialize negotiator.
+        
+        Args:
+            base_url: Server base URL
+        """
+        self.base_url = base_url
+        self._detected_transport: Optional[str] = None
+    
+    @property
+    def detected_transport(self) -> Optional[str]:
+        """Get the detected transport type."""
+        return self._detected_transport
+    
+    def set_detected_transport(self, transport: str) -> None:
+        """
+        Set the detected transport type.
+        
+        Args:
+            transport: Transport type ("http_stream" or "sse")
+        """
+        self._detected_transport = transport
+    
+    def get_endpoint_url(self) -> str:
+        """
+        Get the appropriate endpoint URL based on detected transport.
+        
+        Returns:
+            Endpoint URL for the detected transport
+        """
+        if self._detected_transport == "sse":
+            return get_legacy_sse_endpoint(self.base_url)
+        return get_streamable_http_endpoint(self.base_url)
+    
+    def should_try_streamable_first(self) -> bool:
+        """
+        Determine if Streamable HTTP should be tried first.
+        
+        Per MCP spec, clients should try Streamable HTTP first
+        unless the URL explicitly indicates legacy SSE.
+        
+        Returns:
+            True if should try Streamable HTTP first
+        """
+        return not is_legacy_sse_url(self.base_url)

src/praisonai-agents/praisonaiagents/mcp/mcp_http_stream.py

@@ -200,16 +200,31 @@ class HTTPStreamTransport:
     
     This transport provides a single endpoint for all MCP communication,
     supporting both batch (JSON) and streaming (SSE) response modes.
+    
+    Per MCP Protocol Revision 2025-11-25:
+    - Includes MCP-Protocol-Version header on all requests
+    - Handles session management via Mcp-Session-Id header
+    - Supports SSE resumability via Last-Event-ID
     """
     
+    # Default protocol version for backward compatibility
+    DEFAULT_PROTOCOL_VERSION = '2025-03-26'
+    
     def __init__(self, base_url: str, session_id: Optional[str] = None, options: Optional[Dict[str, Any]] = None):
         self.base_url = base_url
         self.session_id = session_id
         self.options = options or {}
         self.response_mode = self.options.get('responseMode', 'batch')
+        self.protocol_version = self.options.get('protocol_version', self.DEFAULT_PROTOCOL_VERSION)
+        
+        # Track last event ID for resumability
+        self.last_event_id: Optional[str] = None
+        self._retry_delay_ms: int = 3000  # Default retry delay
+        
         self.headers = {
             'Content-Type': 'application/json',
-            'Accept': 'application/json, text/event-stream'
+            'Accept': 'application/json, text/event-stream',
+            'Mcp-Protocol-Version': self.protocol_version  # Required per spec
         }
         if session_id:
             self.headers['Mcp-Session-Id'] = session_id
@@ -281,7 +296,7 @@ class HTTPStreamTransport:
             raise
     
     async def _process_sse_response(self, response):
-        """Process SSE response stream."""
+        """Process SSE response stream with resumability support."""
         buffer = ""
         async for chunk in response.content:
             buffer += chunk.decode('utf-8')
@@ -291,12 +306,34 @@ class HTTPStreamTransport:
                 event, buffer = buffer.split("\n\n", 1)
                 lines = event.strip().split("\n")
                 
-                # Parse SSE event
-                data = None
-                for line in lines:
-                    if line.startswith("data: "):
-                        data = line[6:]  # Remove "data: " prefix
+                # Parse SSE event fields per spec
+                event_id = None
+                data_lines = []
+                retry_value = None
                 
+                for line in lines:
+                    if line.startswith("id:"):
+                        # Track event ID for resumability
+                        event_id = line[3:].strip()
+                    elif line.startswith("data:"):
+                        data_lines.append(line[5:].strip())
+                    elif line.startswith("retry:"):
+                        # Handle retry field per SSE spec
+                        try:
+                            retry_value = int(line[6:].strip())
+                        except ValueError:
+                            pass
+                
+                # Update last event ID for resumability
+                if event_id:
+                    self.last_event_id = event_id
+                
+                # Update retry delay if provided
+                if retry_value is not None:
+                    self._retry_delay_ms = retry_value
+                
+                # Process data if present
+                data = '\n'.join(data_lines) if data_lines else None
                 if data:
                     try:
                         message = json.loads(data)
@@ -310,13 +347,49 @@ class HTTPStreamTransport:
                     except json.JSONDecodeError:
                         logger.error(f"Failed to parse SSE event: {data}")
     
+    async def terminate_session(self) -> bool:
+        """
+        Terminate the session via HTTP DELETE.
+        
+        Per MCP spec: Clients that no longer need a particular session
+        SHOULD send an HTTP DELETE to the MCP endpoint with the
+        Mcp-Session-Id header to explicitly terminate the session.
+        
+        Returns:
+            True if session was terminated, False if server doesn't support it
+        """
+        if not self._session or not self.session_id:
+            return False
+        
+        try:
+            headers = {
+                'Mcp-Session-Id': self.session_id,
+                'Mcp-Protocol-Version': self.protocol_version
+            }
+            async with self._session.delete(self.base_url, headers=headers) as response:
+                if response.status == 405:
+                    # Server doesn't allow client-initiated session termination
+                    logger.debug("Server does not allow session termination via DELETE")
+                    return False
+                elif response.status in (200, 202, 204):
+                    # Session terminated successfully
+                    self.session_id = None
+                    self.headers.pop('Mcp-Session-Id', None)
+                    return True
+                else:
+                    logger.warning(f"Unexpected response to DELETE: {response.status}")
+                    return False
+        except Exception as e:
+            logger.error(f"Error terminating session: {e}")
+            return False
+    
     async def start_sse_listener(self):
         """Start listening for SSE events from the server."""
         if self._sse_task is None or self._sse_task.done():
             self._sse_task = asyncio.create_task(self._sse_listener())
     
     async def _sse_listener(self):
-        """Background task to listen for SSE events."""
+        """Background task to listen for SSE events with resumability support."""
         while True:
             try:
                 # Check if we should stop
@@ -324,18 +397,28 @@ class HTTPStreamTransport:
                     break
                     
                 url = self.base_url
-                if self.session_id:
-                    # Add session as query parameter for SSE connection
-                    url = f"{url}?session={self.session_id}"
                 
+                # Build headers per MCP spec
                 headers = {
                     'Accept': 'text/event-stream',
-                    'Cache-Control': 'no-cache'
+                    'Cache-Control': 'no-cache',
+                    'Mcp-Protocol-Version': self.protocol_version  # Required per spec
                 }
                 if self.session_id:
                     headers['Mcp-Session-Id'] = self.session_id
                 
+                # Include Last-Event-ID for resumability per spec
+                if self.last_event_id:
+                    headers['Last-Event-ID'] = self.last_event_id
+                
                 async with self._session.get(url, headers=headers) as response:
+                    # Handle session expiration (HTTP 404)
+                    if response.status == 404:
+                        logger.warning("Session expired (HTTP 404), need to reinitialize")
+                        self.session_id = None
+                        self.headers.pop('Mcp-Session-Id', None)
+                        break
+                    
                     buffer = ""
                     async for chunk in response.content:
                         # Check if we should stop
@@ -349,12 +432,32 @@ class HTTPStreamTransport:
                             event, buffer = buffer.split("\n\n", 1)
                             lines = event.strip().split("\n")
                             
-                            # Parse SSE event
-                            data = None
-                            for line in lines:
-                                if line.startswith("data: "):
-                                    data = line[6:]  # Remove "data: " prefix
+                            # Parse SSE event fields
+                            event_id = None
+                            data_lines = []
+                            retry_value = None
                             
+                            for line in lines:
+                                if line.startswith("id:"):
+                                    event_id = line[3:].strip()
+                                elif line.startswith("data:"):
+                                    data_lines.append(line[5:].strip())
+                                elif line.startswith("retry:"):
+                                    try:
+                                        retry_value = int(line[6:].strip())
+                                    except ValueError:
+                                        pass
+                            
+                            # Update last event ID for resumability
+                            if event_id:
+                                self.last_event_id = event_id
+                            
+                            # Update retry delay if provided
+                            if retry_value is not None:
+                                self._retry_delay_ms = retry_value
+                            
+                            # Process data
+                            data = '\n'.join(data_lines) if data_lines else None
                             if data:
                                 try:
                                     message = json.loads(data)
@@ -368,7 +471,8 @@ class HTTPStreamTransport:
             except Exception as e:
                 if not (hasattr(self, '_closing') and self._closing):
                     logger.error(f"SSE listener error: {e}")
-                    await asyncio.sleep(1)  # Reconnect after 1 second
+                    # Use retry delay from server or default
+                    await asyncio.sleep(self._retry_delay_ms / 1000.0)
                 else:
                     break
     

src/praisonai-agents/praisonaiagents/mcp/mcp_security.py

@@ -0,0 +1,248 @@
+"""
+MCP Security Module.
+
+This module provides security utilities for MCP transports,
+implementing security best practices from MCP Protocol
+Revision 2025-11-25.
+
+Security features:
+- Origin header validation (DNS rebinding prevention)
+- Localhost binding recommendations
+- Authentication header support
+- Secure session ID generation
+"""
+
+import secrets
+import base64
+from typing import Optional, Dict, List, Set
+from urllib.parse import urlparse
+from dataclasses import dataclass, field
+
+
+# Localhost addresses
+LOCALHOST_ADDRESSES: Set[str] = {
+    'localhost',
+    '127.0.0.1',
+    '::1',  # IPv6 localhost
+}
+
+
+def is_valid_origin(
+    origin: Optional[str],
+    allowed_origins: List[str],
+    allow_missing: bool = False
+) -> bool:
+    """
+    Validate Origin header against allowed origins.
+    
+    Per MCP spec, servers MUST validate the Origin header on all
+    incoming connections to prevent DNS rebinding attacks.
+    
+    Args:
+        origin: The Origin header value
+        allowed_origins: List of allowed origin hosts
+        allow_missing: Whether to allow missing Origin header
+        
+    Returns:
+        True if origin is valid, False otherwise
+    """
+    if not origin:
+        return allow_missing
+    
+    # Extract host from origin
+    host = extract_origin_host(origin)
+    if not host:
+        return False
+    
+    # Check against allowed origins
+    return host in allowed_origins
+
+
+def extract_origin_host(origin: str) -> Optional[str]:
+    """
+    Extract the host from an Origin header value.
+    
+    Args:
+        origin: Origin header value (e.g., "https://example.com:443")
+        
+    Returns:
+        Host portion of the origin, or None if invalid
+    """
+    if not origin:
+        return None
+    
+    try:
+        parsed = urlparse(origin)
+        return parsed.hostname
+    except Exception:
+        return None
+
+
+def is_localhost_address(address: str) -> bool:
+    """
+    Check if an address is a localhost address.
+    
+    Args:
+        address: IP address or hostname
+        
+    Returns:
+        True if localhost, False otherwise
+    """
+    return address.lower() in LOCALHOST_ADDRESSES
+
+
+def should_bind_localhost_only(is_local: bool = True) -> bool:
+    """
+    Determine if server should bind to localhost only.
+    
+    Per MCP spec, when running locally, servers SHOULD bind only
+    to localhost (127.0.0.1) rather than all network interfaces.
+    
+    Args:
+        is_local: Whether this is a local development server
+        
+    Returns:
+        True if should bind to localhost only
+    """
+    return is_local
+
+
+def is_potential_dns_rebinding(origin: str, server_host: str) -> bool:
+    """
+    Detect potential DNS rebinding attack.
+    
+    DNS rebinding occurs when an external origin tries to access
+    a local server through DNS manipulation.
+    
+    Args:
+        origin: The Origin header value
+        server_host: The host the server is bound to
+        
+    Returns:
+        True if potential DNS rebinding attempt
+    """
+    origin_host = extract_origin_host(origin)
+    if not origin_host:
+        return True  # Treat invalid origin as suspicious
+    
+    # If server is on localhost, only localhost origins are safe
+    if is_localhost_address(server_host):
+        return not is_localhost_address(origin_host)
+    
+    # For non-localhost servers, check if origin matches server
+    return origin_host != server_host
+
+
+def create_auth_header(
+    credentials: str,
+    auth_type: str = "bearer",
+    header_name: str = "Authorization"
+) -> Dict[str, str]:
+    """
+    Create an authentication header.
+    
+    Args:
+        credentials: The credentials/token
+        auth_type: Type of auth ("bearer", "basic", "custom")
+        header_name: Custom header name (for auth_type="custom")
+        
+    Returns:
+        Dictionary with authentication header
+    """
+    if auth_type.lower() == "bearer":
+        return {"Authorization": f"Bearer {credentials}"}
+    
+    elif auth_type.lower() == "basic":
+        # Base64 encode for Basic auth
+        encoded = base64.b64encode(credentials.encode()).decode()
+        return {"Authorization": f"Basic {encoded}"}
+    
+    elif auth_type.lower() == "custom":
+        return {header_name: credentials}
+    
+    else:
+        raise ValueError(f"Unknown auth_type: {auth_type}")
+
+
+def validate_auth_header(headers: Dict[str, str]) -> bool:
+    """
+    Validate that authentication header is present.
+    
+    Args:
+        headers: Request headers dictionary
+        
+    Returns:
+        True if Authorization header is present
+    """
+    # Check for Authorization header (case-insensitive)
+    for key in headers:
+        if key.lower() == 'authorization':
+            return bool(headers[key])
+    return False
+
+
+def generate_secure_session_id(length: int = 32) -> str:
+    """
+    Generate a cryptographically secure session ID.
+    
+    Per MCP spec, session IDs SHOULD be globally unique and
+    cryptographically secure (e.g., securely generated UUID,
+    JWT, or cryptographic hash).
+    
+    The ID MUST only contain visible ASCII characters (0x21-0x7E).
+    
+    Args:
+        length: Desired length of session ID (minimum 32)
+        
+    Returns:
+        Secure session ID string
+    """
+    # Ensure minimum length for security
+    length = max(length, 32)
+    
+    # Use URL-safe base64 which only uses visible ASCII
+    # Generate extra bytes to account for base64 expansion
+    num_bytes = (length * 3) // 4 + 1
+    random_bytes = secrets.token_bytes(num_bytes)
+    
+    # Encode to URL-safe base64 (uses A-Z, a-z, 0-9, -, _)
+    session_id = base64.urlsafe_b64encode(random_bytes).decode()
+    
+    # Trim to desired length
+    return session_id[:length]
+
+
+@dataclass
+class SecurityConfig:
+    """
+    Security configuration for MCP transports.
+    
+    This class holds security settings that can be applied
+    to MCP server implementations.
+    
+    Attributes:
+        validate_origin: Whether to validate Origin header
+        allow_missing_origin: Whether to allow missing Origin
+        allowed_origins: List of allowed origin hosts
+        require_auth: Whether authentication is required
+        bind_localhost_only: Whether to bind to localhost only
+    """
+    validate_origin: bool = True
+    allow_missing_origin: bool = False
+    allowed_origins: List[str] = field(default_factory=lambda: ["localhost", "127.0.0.1"])
+    require_auth: bool = False
+    bind_localhost_only: bool = True
+    
+    def is_origin_allowed(self, origin: Optional[str]) -> bool:
+        """Check if an origin is allowed by this config."""
+        if not self.validate_origin:
+            return True
+        return is_valid_origin(
+            origin,
+            self.allowed_origins,
+            self.allow_missing_origin
+        )
+    
+    def get_bind_address(self) -> str:
+        """Get recommended bind address based on config."""
+        return "127.0.0.1" if self.bind_localhost_only else "0.0.0.0"

src/praisonai-agents/praisonaiagents/mcp/mcp_server.py

@@ -0,0 +1,356 @@
+"""MCP Server wrapper for exposing PraisonAI tools as MCP servers.
+
+This module provides a simple way to expose Python functions as MCP tools
+that can be consumed by any MCP client (Claude Desktop, Cursor, etc.).
+
+Usage:
+    from praisonaiagents.mcp import ToolsMCPServer
+    
+    def search(query: str) -> str:
+        '''Search the web.'''
+        return f"Results for {query}"
+    
+    server = ToolsMCPServer(name="my-tools")
+    server.register_tool(search)
+    server.run()  # Starts MCP server (stdio by default)
+"""
+
+import asyncio
+import inspect
+import logging
+from typing import Any, Callable, Dict, List, Optional
+
+from .mcp_utils import function_to_mcp_schema
+
+logger = logging.getLogger(__name__)
+
+
+class ToolsMCPServer:
+    """MCP Server that exposes Python functions as MCP tools.
+    
+    This class wraps Python functions and exposes them as MCP tools
+    using the FastMCP library. It supports both sync and async functions.
+    
+    Example:
+        server = ToolsMCPServer(name="my-tools")
+        
+        def search(query: str) -> str:
+            '''Search the web.'''
+            return f"Results for {query}"
+        
+        server.register_tool(search)
+        server.run()  # Starts stdio server
+    """
+    
+    def __init__(
+        self,
+        name: str = "praisonai-tools",
+        tools: Optional[List[Callable]] = None,
+        debug: bool = False
+    ):
+        """Initialize the MCP server.
+        
+        Args:
+            name: Name of the MCP server
+            tools: Optional list of tools to register on init
+            debug: Enable debug logging
+        """
+        self.name = name
+        self.tools: List[Callable] = []
+        self._tool_map: Dict[str, Callable] = {}
+        self._fastmcp = None
+        self._debug = debug
+        
+        if debug:
+            logger.setLevel(logging.DEBUG)
+        
+        # Register initial tools if provided
+        if tools:
+            self.register_tools(tools)
+    
+    def register_tool(self, func: Callable) -> None:
+        """Register a single tool function.
+        
+        Args:
+            func: The function to register as an MCP tool
+        """
+        if not callable(func):
+            raise TypeError(f"Expected callable, got {type(func)}")
+        
+        self.tools.append(func)
+        tool_name = getattr(func, '__mcp_name__', None) or func.__name__
+        self._tool_map[tool_name] = func
+        
+        logger.debug(f"Registered tool: {tool_name}")
+    
+    def register_tools(self, funcs: List[Callable]) -> None:
+        """Register multiple tool functions.
+        
+        Args:
+            funcs: List of functions to register as MCP tools
+        """
+        for func in funcs:
+            self.register_tool(func)
+    
+    def get_tool_schemas(self) -> List[Dict[str, Any]]:
+        """Get MCP schemas for all registered tools.
+        
+        Returns:
+            List of MCP tool schema dictionaries
+        """
+        return [function_to_mcp_schema(tool) for tool in self.tools]
+    
+    def get_tool_names(self) -> List[str]:
+        """Get names of all registered tools.
+        
+        Returns:
+            List of tool names
+        """
+        return list(self._tool_map.keys())
+    
+    def execute_tool(self, tool_name: str, arguments: Dict[str, Any]) -> Any:
+        """Execute a registered tool synchronously.
+        
+        Args:
+            tool_name: Name of the tool to execute
+            arguments: Arguments to pass to the tool
+        
+        Returns:
+            Result from the tool execution
+        
+        Raises:
+            ValueError: If tool is not found
+        """
+        if tool_name not in self._tool_map:
+            raise ValueError(f"Tool '{tool_name}' not found")
+        
+        func = self._tool_map[tool_name]
+        
+        # Handle async functions
+        if asyncio.iscoroutinefunction(func):
+            loop = asyncio.new_event_loop()
+            try:
+                return loop.run_until_complete(func(**arguments))
+            finally:
+                loop.close()
+        
+        return func(**arguments)
+    
+    async def execute_tool_async(self, tool_name: str, arguments: Dict[str, Any]) -> Any:
+        """Execute a registered tool asynchronously.
+        
+        Args:
+            tool_name: Name of the tool to execute
+            arguments: Arguments to pass to the tool
+        
+        Returns:
+            Result from the tool execution
+        
+        Raises:
+            ValueError: If tool is not found
+        """
+        if tool_name not in self._tool_map:
+            raise ValueError(f"Tool '{tool_name}' not found")
+        
+        func = self._tool_map[tool_name]
+        
+        if asyncio.iscoroutinefunction(func):
+            return await func(**arguments)
+        
+        # Run sync function in executor
+        loop = asyncio.get_event_loop()
+        return await loop.run_in_executor(None, lambda: func(**arguments))
+    
+    def get_fastmcp(self):
+        """Get or create FastMCP instance with registered tools.
+        
+        Returns:
+            FastMCP instance with all tools registered
+        """
+        if self._fastmcp is None:
+            self._fastmcp = self._create_fastmcp()
+        return self._fastmcp
+    
+    def _create_fastmcp(self):
+        """Create FastMCP instance and register all tools.
+        
+        Returns:
+            Configured FastMCP instance
+        """
+        try:
+            from mcp.server.fastmcp import FastMCP
+        except ImportError:
+            raise ImportError(
+                "FastMCP not available. Install with: pip install mcp"
+            )
+        
+        mcp = FastMCP(self.name)
+        
+        # Register each tool with FastMCP
+        for func in self.tools:
+            self._register_tool_with_fastmcp(mcp, func)
+        
+        return mcp
+    
+    def _register_tool_with_fastmcp(self, mcp, func: Callable) -> None:
+        """Register a single function with FastMCP.
+        
+        Args:
+            mcp: FastMCP instance
+            func: Function to register
+        """
+        # Get tool metadata
+        tool_name = getattr(func, '__mcp_name__', None) or func.__name__
+        description = getattr(func, '__mcp_description__', None)
+        if description is None:
+            docstring = inspect.getdoc(func)
+            description = docstring.split('\n')[0] if docstring else tool_name
+        
+        # Create wrapper that preserves signature
+        if asyncio.iscoroutinefunction(func):
+            @mcp.tool(name=tool_name, description=description)
+            async def tool_wrapper(**kwargs):
+                return await func(**kwargs)
+        else:
+            @mcp.tool(name=tool_name, description=description)
+            def tool_wrapper(**kwargs):
+                return func(**kwargs)
+        
+        # Copy signature for proper schema generation
+        tool_wrapper.__signature__ = inspect.signature(func)
+        tool_wrapper.__annotations__ = getattr(func, '__annotations__', {})
+    
+    def run(self, transport: str = "stdio") -> None:
+        """Run the MCP server.
+        
+        Args:
+            transport: Transport type - "stdio" or "sse"
+        """
+        if transport == "stdio":
+            self.run_stdio()
+        elif transport == "sse":
+            self.run_sse()
+        else:
+            raise ValueError(f"Unknown transport: {transport}. Use 'stdio' or 'sse'")
+    
+    def run_stdio(self) -> None:
+        """Run the MCP server using stdio transport."""
+        mcp = self.get_fastmcp()
+        
+        print(f"🚀 Starting MCP server '{self.name}' with stdio transport")
+        print(f"🛠️  Available tools: {', '.join(self.get_tool_names())}")
+        
+        mcp.run()
+    
+    def run_sse(self, host: str = "0.0.0.0", port: int = 8080) -> None:
+        """Run the MCP server using SSE transport.
+        
+        Args:
+            host: Host to bind to
+            port: Port to listen on
+        """
+        try:
+            import uvicorn
+            from mcp.server.sse import SseServerTransport
+            from starlette.applications import Starlette
+            from starlette.requests import Request
+            from starlette.routing import Mount, Route
+        except ImportError as e:
+            raise ImportError(
+                f"SSE dependencies not available: {e}. "
+                "Install with: pip install uvicorn starlette"
+            )
+        
+        mcp = self.get_fastmcp()
+        
+        # Set up SSE transport
+        sse_path = "/sse"
+        messages_path = "/messages/"
+        
+        sse_transport = SseServerTransport(messages_path)
+        
+        async def handle_sse(request: Request):
+            async with sse_transport.connect_sse(
+                request.scope, request.receive, request._send
+            ) as (read_stream, write_stream):
+                await mcp._mcp_server.run(
+                    read_stream,
+                    write_stream,
+                    mcp._mcp_server.create_initialization_options()
+                )
+        
+        app = Starlette(
+            debug=self._debug,
+            routes=[
+                Route(sse_path, endpoint=handle_sse),
+                Mount(messages_path, app=sse_transport.handle_post_message),
+            ]
+        )
+        
+        print(f"🚀 Starting MCP server '{self.name}' with SSE transport")
+        print(f"📡 SSE endpoint: http://{host}:{port}{sse_path}")
+        print(f"🛠️  Available tools: {', '.join(self.get_tool_names())}")
+        
+        uvicorn.run(app, host=host, port=port)
+
+
+def launch_tools_mcp_server(
+    tools: Optional[List[Callable]] = None,
+    tool_names: Optional[List[str]] = None,
+    name: str = "praisonai-tools",
+    transport: str = "stdio",
+    host: str = "0.0.0.0",
+    port: int = 8080,
+    debug: bool = False
+) -> None:
+    """Convenience function to launch an MCP server with tools.
+    
+    Args:
+        tools: List of tool functions to expose
+        tool_names: List of tool names to load from praisonaiagents.tools
+        name: Name of the MCP server
+        transport: Transport type - "stdio" or "sse"
+        host: Host for SSE transport
+        port: Port for SSE transport
+        debug: Enable debug logging
+    
+    Example:
+        # With custom tools
+        def my_search(query: str) -> str:
+            return f"Results for {query}"
+        
+        launch_tools_mcp_server(tools=[my_search])
+        
+        # With built-in tools
+        launch_tools_mcp_server(tool_names=["tavily_search", "exa_search"])
+    """
+    server = ToolsMCPServer(name=name, debug=debug)
+    
+    # Register custom tools
+    if tools:
+        server.register_tools(tools)
+    
+    # Load and register tools by name
+    if tool_names:
+        from praisonaiagents import tools as praison_tools
+        
+        for tool_name in tool_names:
+            if hasattr(praison_tools, tool_name):
+                tool_func = getattr(praison_tools, tool_name)
+                if callable(tool_func):
+                    server.register_tool(tool_func)
+                else:
+                    logger.warning(f"Tool '{tool_name}' is not callable")
+            else:
+                logger.warning(f"Tool '{tool_name}' not found in praisonaiagents.tools")
+    
+    if not server.tools:
+        raise ValueError("No tools registered. Provide 'tools' or 'tool_names'.")
+    
+    # Run the server
+    if transport == "stdio":
+        server.run_stdio()
+    elif transport == "sse":
+        server.run_sse(host=host, port=port)
+    else:
+        raise ValueError(f"Unknown transport: {transport}")

src/praisonai-agents/praisonaiagents/mcp/mcp_session.py

@@ -0,0 +1,457 @@
+"""
+MCP Session Management Module.
+
+This module provides session management utilities for MCP transports,
+implementing the session management requirements from MCP Protocol
+Revision 2025-11-25.
+
+Features:
+- Session ID validation (visible ASCII 0x21-0x7E)
+- Session lifecycle management
+- Protocol version header handling
+- Session expiration detection
+"""
+
+import re
+from typing import Optional, Dict, Any
+
+
+# Valid protocol versions
+VALID_PROTOCOL_VERSIONS = {
+    '2024-11-05',  # Original HTTP+SSE
+    '2025-03-26',  # Streamable HTTP introduction
+    '2025-06-18',  # Updates
+    '2025-11-25',  # Current revision
+}
+
+# Default protocol version for backward compatibility
+DEFAULT_PROTOCOL_VERSION = '2025-03-26'
+
+
+def is_valid_session_id(session_id: Optional[str]) -> bool:
+    """
+    Validate MCP session ID per specification.
+    
+    Session IDs MUST only contain visible ASCII characters (0x21 to 0x7E).
+    This excludes space (0x20), control characters (0x00-0x1F), and DEL (0x7F).
+    
+    Args:
+        session_id: The session ID to validate
+        
+    Returns:
+        True if valid, False otherwise
+    """
+    if session_id is None or session_id == "":
+        return False
+    
+    # Check each character is in visible ASCII range (0x21-0x7E)
+    for char in session_id:
+        code = ord(char)
+        if code < 0x21 or code > 0x7E:
+            return False
+    
+    return True
+
+
+def is_valid_protocol_version(version: Optional[str]) -> bool:
+    """
+    Validate MCP protocol version string.
+    
+    Args:
+        version: Protocol version string (e.g., '2025-11-25')
+        
+    Returns:
+        True if valid, False otherwise
+    """
+    if not version:
+        return False
+    
+    # Check format: YYYY-MM-DD
+    if not re.match(r'^\d{4}-\d{2}-\d{2}$', version):
+        return False
+    
+    return True
+
+
+def is_session_expired(status_code: int) -> bool:
+    """
+    Check if HTTP status code indicates session expiration.
+    
+    Per MCP spec, HTTP 404 indicates the session has been terminated.
+    
+    Args:
+        status_code: HTTP response status code
+        
+    Returns:
+        True if session is expired, False otherwise
+    """
+    return status_code == 404
+
+
+def extract_session_id(headers: Dict[str, str]) -> Optional[str]:
+    """
+    Extract session ID from HTTP response headers.
+    
+    The header name is case-insensitive per HTTP spec.
+    
+    Args:
+        headers: HTTP response headers dictionary
+        
+    Returns:
+        Session ID if present, None otherwise
+    """
+    # Check various case combinations
+    for key in headers:
+        if key.lower() == 'mcp-session-id':
+            return headers[key]
+    return None
+
+
+class SessionManager:
+    """
+    Manages MCP session state for HTTP transports.
+    
+    This class handles:
+    - Session ID storage and validation
+    - Protocol version negotiation
+    - Session lifecycle (active, expired, terminated)
+    - Header generation for requests
+    
+    Example:
+        ```python
+        manager = SessionManager()
+        
+        # After receiving InitializeResult with session ID
+        manager.set_session_id("abc123")
+        
+        # Get headers for subsequent requests
+        headers = manager.get_headers()
+        # {'Mcp-Session-Id': 'abc123', 'Mcp-Protocol-Version': '2025-03-26'}
+        ```
+    """
+    
+    def __init__(self, protocol_version: Optional[str] = None):
+        """
+        Initialize session manager.
+        
+        Args:
+            protocol_version: Initial protocol version (defaults to 2025-03-26)
+        """
+        self._session_id: Optional[str] = None
+        self._protocol_version: str = protocol_version or DEFAULT_PROTOCOL_VERSION
+        self._is_expired: bool = False
+        self._metadata: Dict[str, Any] = {}
+    
+    @property
+    def session_id(self) -> Optional[str]:
+        """Get current session ID."""
+        return self._session_id
+    
+    @property
+    def protocol_version(self) -> str:
+        """Get current protocol version."""
+        return self._protocol_version
+    
+    @property
+    def is_active(self) -> bool:
+        """Check if session is active (has ID and not expired)."""
+        return self._session_id is not None and not self._is_expired
+    
+    @property
+    def is_expired(self) -> bool:
+        """Check if session has expired."""
+        return self._is_expired
+    
+    def set_session_id(self, session_id: str) -> None:
+        """
+        Set the session ID.
+        
+        Args:
+            session_id: The session ID from server
+            
+        Raises:
+            ValueError: If session ID is invalid
+        """
+        if not is_valid_session_id(session_id):
+            raise ValueError("Invalid session ID: must contain only visible ASCII (0x21-0x7E)")
+        
+        self._session_id = session_id
+        self._is_expired = False
+    
+    def set_protocol_version(self, version: str) -> None:
+        """
+        Set the negotiated protocol version.
+        
+        Args:
+            version: Protocol version string
+            
+        Raises:
+            ValueError: If version format is invalid
+        """
+        if not is_valid_protocol_version(version):
+            raise ValueError(f"Invalid protocol version format: {version}")
+        
+        self._protocol_version = version
+    
+    def clear(self) -> None:
+        """Clear the session (logout/disconnect)."""
+        self._session_id = None
+        self._is_expired = False
+        self._metadata = {}
+    
+    def mark_expired(self) -> None:
+        """Mark the session as expired (received HTTP 404)."""
+        self._is_expired = True
+    
+    def get_headers(self) -> Dict[str, str]:
+        """
+        Get HTTP headers for MCP requests.
+        
+        Returns:
+            Dictionary of headers to include in requests
+        """
+        headers = {
+            'Mcp-Protocol-Version': self._protocol_version
+        }
+        
+        if self._session_id:
+            headers['Mcp-Session-Id'] = self._session_id
+        
+        return headers
+    
+    def update_from_response(self, headers: Dict[str, str], status_code: int = 200) -> None:
+        """
+        Update session state from HTTP response.
+        
+        Args:
+            headers: Response headers
+            status_code: HTTP status code
+        """
+        # Check for session expiration
+        if is_session_expired(status_code):
+            self.mark_expired()
+            return
+        
+        # Extract and update session ID if present
+        new_session_id = extract_session_id(headers)
+        if new_session_id:
+            self.set_session_id(new_session_id)
+    
+    def set_metadata(self, key: str, value: Any) -> None:
+        """Store arbitrary metadata with the session."""
+        self._metadata[key] = value
+    
+    def get_metadata(self, key: str, default: Any = None) -> Any:
+        """Retrieve metadata from the session."""
+        return self._metadata.get(key, default)
+
+
+class ResumabilityManager:
+    """
+    Manages SSE stream resumability for MCP transports.
+    
+    This class handles:
+    - Event ID tracking per stream
+    - Last-Event-ID header generation
+    - Retry delay handling
+    
+    Per MCP spec, event IDs should be assigned per-stream to act as
+    a cursor within that particular stream.
+    """
+    
+    def __init__(self):
+        """Initialize resumability manager."""
+        self._last_event_ids: Dict[str, str] = {}  # stream_id -> last_event_id
+        self._retry_delay: int = 3000  # Default retry delay in ms
+    
+    def set_last_event_id(self, stream_id: str, event_id: str) -> None:
+        """
+        Store the last received event ID for a stream.
+        
+        Args:
+            stream_id: Identifier for the SSE stream
+            event_id: The event ID from the SSE event
+        """
+        self._last_event_ids[stream_id] = event_id
+    
+    def get_last_event_id(self, stream_id: str) -> Optional[str]:
+        """
+        Get the last received event ID for a stream.
+        
+        Args:
+            stream_id: Identifier for the SSE stream
+            
+        Returns:
+            Last event ID if available, None otherwise
+        """
+        return self._last_event_ids.get(stream_id)
+    
+    def set_retry_delay(self, delay_ms: int) -> None:
+        """
+        Set the retry delay from SSE retry field.
+        
+        Args:
+            delay_ms: Delay in milliseconds
+        """
+        self._retry_delay = delay_ms
+    
+    def get_retry_delay(self) -> int:
+        """Get current retry delay in milliseconds."""
+        return self._retry_delay
+    
+    def get_resume_headers(self, stream_id: str) -> Dict[str, str]:
+        """
+        Get headers for resuming a stream.
+        
+        Args:
+            stream_id: Identifier for the SSE stream
+            
+        Returns:
+            Headers dict with Last-Event-ID if available
+        """
+        headers = {}
+        last_event_id = self.get_last_event_id(stream_id)
+        if last_event_id:
+            headers['Last-Event-ID'] = last_event_id
+        return headers
+    
+    def clear_stream(self, stream_id: str) -> None:
+        """Clear tracking for a specific stream."""
+        self._last_event_ids.pop(stream_id, None)
+    
+    def clear_all(self) -> None:
+        """Clear all stream tracking."""
+        self._last_event_ids.clear()
+
+
+def parse_sse_event(event_text: str) -> Dict[str, Any]:
+    """
+    Parse an SSE event into its components.
+    
+    SSE events have the format:
+        id: <event-id>
+        retry: <milliseconds>
+        data: <data-line>
+        data: <data-line-2>
+        
+        (blank line ends event)
+    
+    Args:
+        event_text: Raw SSE event text
+        
+    Returns:
+        Dictionary with 'id', 'retry', 'data', 'event' fields as available
+    """
+    result: Dict[str, Any] = {}
+    data_lines = []
+    
+    for line in event_text.strip().split('\n'):
+        if line.startswith('id:'):
+            result['id'] = line[3:].strip()
+        elif line.startswith('retry:'):
+            try:
+                result['retry'] = int(line[6:].strip())
+            except ValueError:
+                pass
+        elif line.startswith('data:'):
+            data_lines.append(line[5:].strip())
+        elif line.startswith('event:'):
+            result['event'] = line[6:].strip()
+    
+    # Join multiple data lines with newlines
+    if data_lines:
+        result['data'] = '\n'.join(data_lines)
+    
+    return result
+
+
+class EventStore:
+    """
+    Optional event store for message redelivery support.
+    
+    This class stores events per stream to enable replay when
+    clients reconnect with Last-Event-ID header.
+    
+    Note: This is an in-memory store suitable for single-server
+    deployments. For multi-node deployments, use an external
+    store (Redis, database, etc.).
+    """
+    
+    def __init__(self, max_events: int = 1000):
+        """
+        Initialize event store.
+        
+        Args:
+            max_events: Maximum events to store per stream
+        """
+        self.max_events = max_events
+        self._streams: Dict[str, list] = {}  # stream_id -> [(event_id, data), ...]
+    
+    def store(self, stream_id: str, event_id: str, data: Any) -> None:
+        """
+        Store an event for potential replay.
+        
+        Args:
+            stream_id: Identifier for the SSE stream
+            event_id: The event ID
+            data: The event data
+        """
+        if stream_id not in self._streams:
+            self._streams[stream_id] = []
+        
+        self._streams[stream_id].append((event_id, data))
+        
+        # Trim to max_events
+        if len(self._streams[stream_id]) > self.max_events:
+            self._streams[stream_id] = self._streams[stream_id][-self.max_events:]
+    
+    def get_events_after(self, stream_id: str, after_event_id: str) -> list:
+        """
+        Get events after a specific event ID.
+        
+        Args:
+            stream_id: Identifier for the SSE stream
+            after_event_id: The Last-Event-ID from client
+            
+        Returns:
+            List of event data after the specified event
+        """
+        if stream_id not in self._streams:
+            return []
+        
+        events = self._streams[stream_id]
+        
+        # Find the index of the after_event_id
+        start_index = None
+        for i, (event_id, _) in enumerate(events):
+            if event_id == after_event_id:
+                start_index = i + 1
+                break
+        
+        if start_index is None:
+            return []
+        
+        return [data for _, data in events[start_index:]]
+    
+    def get_all_events(self, stream_id: str) -> list:
+        """
+        Get all stored events for a stream.
+        
+        Args:
+            stream_id: Identifier for the SSE stream
+            
+        Returns:
+            List of all event data
+        """
+        if stream_id not in self._streams:
+            return []
+        
+        return [data for _, data in self._streams[stream_id]]
+    
+    def clear_stream(self, stream_id: str) -> None:
+        """Clear events for a specific stream."""
+        self._streams.pop(stream_id, None)
+    
+    def clear_all(self) -> None:
+        """Clear all stored events."""
+        self._streams.clear()

src/praisonai-agents/praisonaiagents/mcp/mcp_transport.py

@@ -0,0 +1,298 @@
+"""
+MCP Transport Abstraction Layer.
+
+This module provides a common interface for all MCP transports,
+enabling easy switching between transport types and custom
+transport implementations.
+
+Transport types:
+- stdio: Standard input/output (subprocess)
+- sse: Server-Sent Events (legacy HTTP+SSE)
+- http_stream: Streamable HTTP (current standard)
+- websocket: WebSocket (SEP-1288)
+"""
+
+import re
+from abc import ABC, abstractmethod
+from typing import Any, Dict, Optional, List, Type
+from dataclasses import dataclass, field
+
+
+@dataclass
+class TransportConfig:
+    """
+    Configuration for MCP transports.
+    
+    Attributes:
+        timeout: Connection/operation timeout in seconds
+        debug: Enable debug logging
+        retry_count: Number of retry attempts
+        retry_delay: Base delay between retries in seconds
+        auth_token: Optional authentication token
+        headers: Additional headers for HTTP transports
+    """
+    timeout: int = 60
+    debug: bool = False
+    retry_count: int = 3
+    retry_delay: float = 1.0
+    auth_token: Optional[str] = None
+    headers: Dict[str, str] = field(default_factory=dict)
+
+
+class BaseTransport(ABC):
+    """
+    Abstract base class for MCP transports.
+    
+    All MCP transports must implement this interface to ensure
+    consistent behavior across transport types.
+    
+    The transport is responsible for:
+    - Establishing and maintaining connections
+    - Sending JSON-RPC messages
+    - Receiving JSON-RPC messages
+    - Proper cleanup on close
+    """
+    
+    @abstractmethod
+    async def connect(self) -> None:
+        """
+        Establish connection to the MCP server.
+        
+        Raises:
+            ConnectionError: If connection fails
+        """
+        pass
+    
+    @abstractmethod
+    async def send(self, message: Dict[str, Any]) -> None:
+        """
+        Send a JSON-RPC message to the server.
+        
+        Args:
+            message: JSON-RPC message dictionary
+            
+        Raises:
+            RuntimeError: If not connected
+        """
+        pass
+    
+    @abstractmethod
+    async def receive(self) -> Dict[str, Any]:
+        """
+        Receive a JSON-RPC message from the server.
+        
+        Returns:
+            JSON-RPC message dictionary
+            
+        Raises:
+            RuntimeError: If not connected
+        """
+        pass
+    
+    @abstractmethod
+    async def close(self) -> None:
+        """
+        Close the connection and cleanup resources.
+        """
+        pass
+    
+    @property
+    @abstractmethod
+    def is_connected(self) -> bool:
+        """Check if transport is currently connected."""
+        pass
+    
+    async def __aenter__(self):
+        """Async context manager entry."""
+        await self.connect()
+        return self
+    
+    async def __aexit__(self, exc_type, exc_val, exc_tb):
+        """Async context manager exit."""
+        await self.close()
+
+
+def get_transport_type(target: str) -> str:
+    """
+    Determine the appropriate transport type for a target.
+    
+    Args:
+        target: URL or command string
+        
+    Returns:
+        Transport type string: "stdio", "sse", "http_stream", or "websocket"
+    """
+    if not isinstance(target, str):
+        return "stdio"
+    
+    # WebSocket URLs
+    if re.match(r'^wss?://', target, re.IGNORECASE):
+        return "websocket"
+    
+    # HTTP URLs
+    if re.match(r'^https?://', target, re.IGNORECASE):
+        # Legacy SSE endpoint
+        if target.endswith('/sse'):
+            return "sse"
+        # Default to Streamable HTTP
+        return "http_stream"
+    
+    # Everything else is stdio (command)
+    return "stdio"
+
+
+class TransportRegistry:
+    """
+    Registry for MCP transport implementations.
+    
+    This allows registration of custom transports and provides
+    a central place to look up transport classes by name.
+    """
+    
+    def __init__(self):
+        """Initialize empty registry."""
+        self._transports: Dict[str, Type[BaseTransport]] = {}
+    
+    def register(self, name: str, transport_class: Type[BaseTransport]) -> None:
+        """
+        Register a transport class.
+        
+        Args:
+            name: Transport name (e.g., "websocket")
+            transport_class: Transport class (must extend BaseTransport)
+        """
+        if not issubclass(transport_class, BaseTransport):
+            raise TypeError(f"{transport_class} must extend BaseTransport")
+        self._transports[name] = transport_class
+    
+    def get(self, name: str) -> Optional[Type[BaseTransport]]:
+        """
+        Get a registered transport class.
+        
+        Args:
+            name: Transport name
+            
+        Returns:
+            Transport class or None if not found
+        """
+        return self._transports.get(name)
+    
+    def list_transports(self) -> List[str]:
+        """
+        List all registered transport names.
+        
+        Returns:
+            List of transport names
+        """
+        return list(self._transports.keys())
+    
+    def unregister(self, name: str) -> None:
+        """
+        Unregister a transport.
+        
+        Args:
+            name: Transport name to remove
+        """
+        self._transports.pop(name, None)
+
+
+# Placeholder transport classes for registry
+# These are minimal implementations to satisfy the registry
+
+class StdioTransportPlaceholder(BaseTransport):
+    """Placeholder for stdio transport."""
+    async def connect(self): pass
+    async def send(self, message): pass
+    async def receive(self): return {}
+    async def close(self): pass
+    @property
+    def is_connected(self): return False
+
+
+class SSETransportPlaceholder(BaseTransport):
+    """Placeholder for SSE transport."""
+    async def connect(self): pass
+    async def send(self, message): pass
+    async def receive(self): return {}
+    async def close(self): pass
+    @property
+    def is_connected(self): return False
+
+
+class HTTPStreamTransportPlaceholder(BaseTransport):
+    """Placeholder for HTTP Stream transport."""
+    async def connect(self): pass
+    async def send(self, message): pass
+    async def receive(self): return {}
+    async def close(self): pass
+    @property
+    def is_connected(self): return False
+
+
+class WebSocketTransportPlaceholder(BaseTransport):
+    """Placeholder for WebSocket transport."""
+    async def connect(self): pass
+    async def send(self, message): pass
+    async def receive(self): return {}
+    async def close(self): pass
+    @property
+    def is_connected(self): return False
+
+
+# Default registry instance
+_default_registry: Optional[TransportRegistry] = None
+
+
+def get_default_registry() -> TransportRegistry:
+    """
+    Get the default transport registry with built-in transports.
+    
+    Returns:
+        TransportRegistry with default transports registered
+    """
+    global _default_registry
+    
+    if _default_registry is None:
+        _default_registry = TransportRegistry()
+        
+        # Register default transports
+        _default_registry.register("stdio", StdioTransportPlaceholder)
+        _default_registry.register("sse", SSETransportPlaceholder)
+        _default_registry.register("http_stream", HTTPStreamTransportPlaceholder)
+        _default_registry.register("websocket", WebSocketTransportPlaceholder)
+    
+    return _default_registry
+
+
+def create_transport(
+    target: str,
+    config: Optional[TransportConfig] = None,
+    registry: Optional[TransportRegistry] = None
+) -> BaseTransport:
+    """
+    Create a transport instance for the given target.
+    
+    Args:
+        target: URL or command string
+        config: Optional transport configuration
+        registry: Optional custom registry (uses default if None)
+        
+    Returns:
+        Transport instance
+        
+    Raises:
+        ValueError: If no suitable transport found
+    """
+    if registry is None:
+        registry = get_default_registry()
+    
+    if config is None:
+        config = TransportConfig()
+    
+    transport_type = get_transport_type(target)
+    transport_class = registry.get(transport_type)
+    
+    if transport_class is None:
+        raise ValueError(f"No transport registered for type: {transport_type}")
+    
+    return transport_class()

src/praisonai-agents/praisonaiagents/mcp/mcp_utils.py

@@ -0,0 +1,238 @@
+"""MCP utility functions for converting Python functions to MCP tool schemas.
+
+This module provides utilities to convert Python functions with type hints
+into MCP-compatible tool schemas for exposing as MCP server tools.
+
+Usage:
+    from praisonaiagents.mcp.mcp_utils import function_to_mcp_schema
+    
+    def my_tool(query: str, max_results: int = 10) -> str:
+        '''Search for information.'''
+        return "results"
+    
+    schema = function_to_mcp_schema(my_tool)
+    # Returns MCP-compatible tool schema
+"""
+
+import inspect
+from typing import Any, Callable, Dict, List, Union, get_type_hints, get_origin, get_args
+
+
+def python_type_to_json_schema(python_type: Any) -> Dict[str, Any]:
+    """Convert a Python type annotation to JSON Schema.
+    
+    Args:
+        python_type: A Python type annotation (e.g., str, int, List[str], Optional[int])
+    
+    Returns:
+        JSON Schema dictionary representing the type
+    """
+    # Handle None type
+    if python_type is None or python_type is type(None):
+        return {"type": "null"}
+    
+    # Handle Any type
+    if python_type is Any:
+        return {}
+    
+    # Handle basic types
+    type_mapping = {
+        str: {"type": "string"},
+        int: {"type": "integer"},
+        float: {"type": "number"},
+        bool: {"type": "boolean"},
+        bytes: {"type": "string"},
+    }
+    
+    if python_type in type_mapping:
+        return type_mapping[python_type]
+    
+    # Handle generic types (List, Dict, Optional, Union)
+    origin = get_origin(python_type)
+    args = get_args(python_type)
+    
+    # Handle List[T]
+    if origin is list:
+        if args:
+            return {
+                "type": "array",
+                "items": python_type_to_json_schema(args[0])
+            }
+        return {"type": "array", "items": {}}
+    
+    # Handle Dict[K, V]
+    if origin is dict:
+        return {"type": "object"}
+    
+    # Handle Optional[T] (Union[T, None])
+    if origin is Union:
+        # Filter out None type
+        non_none_args = [arg for arg in args if arg is not type(None)]
+        if len(non_none_args) == 1:
+            # Optional[T] case
+            return python_type_to_json_schema(non_none_args[0])
+        elif len(non_none_args) > 1:
+            # Union of multiple types
+            return {"oneOf": [python_type_to_json_schema(arg) for arg in non_none_args]}
+        return {}
+    
+    # Handle tuple
+    if origin is tuple:
+        if args:
+            return {
+                "type": "array",
+                "items": [python_type_to_json_schema(arg) for arg in args]
+            }
+        return {"type": "array"}
+    
+    # Handle set
+    if origin is set:
+        if args:
+            return {
+                "type": "array",
+                "items": python_type_to_json_schema(args[0]),
+                "uniqueItems": True
+            }
+        return {"type": "array", "uniqueItems": True}
+    
+    # Default to object for unknown types
+    return {"type": "object"}
+
+
+def get_tool_metadata(func: Callable) -> Dict[str, str]:
+    """Extract metadata from a function for MCP tool registration.
+    
+    Checks for custom attributes (__mcp_name__, __mcp_description__) first,
+    then falls back to function name and docstring.
+    
+    Args:
+        func: The function to extract metadata from
+    
+    Returns:
+        Dictionary with 'name' and 'description' keys
+    """
+    # Get name - check for custom attribute first
+    name = getattr(func, '__mcp_name__', None) or func.__name__
+    
+    # Get description - check for custom attribute first
+    description = getattr(func, '__mcp_description__', None)
+    
+    if description is None:
+        # Extract from docstring
+        docstring = inspect.getdoc(func)
+        if docstring:
+            # Use first line of docstring as description
+            description = docstring.split('\n')[0].strip()
+        else:
+            # Fall back to function name
+            description = name
+    
+    return {
+        "name": name,
+        "description": description
+    }
+
+
+def function_to_mcp_schema(func: Callable) -> Dict[str, Any]:
+    """Convert a Python function to an MCP tool schema.
+    
+    This function inspects the function's signature, type hints, and docstring
+    to generate a complete MCP-compatible tool schema.
+    
+    Args:
+        func: The function to convert
+    
+    Returns:
+        MCP tool schema dictionary with:
+        - name: Tool name
+        - description: Tool description
+        - inputSchema: JSON Schema for parameters
+    
+    Example:
+        def search(query: str, max_results: int = 10) -> str:
+            '''Search the web for information.'''
+            return "results"
+        
+        schema = function_to_mcp_schema(search)
+        # {
+        #     "name": "search",
+        #     "description": "Search the web for information.",
+        #     "inputSchema": {
+        #         "type": "object",
+        #         "properties": {
+        #             "query": {"type": "string"},
+        #             "max_results": {"type": "integer"}
+        #         },
+        #         "required": ["query"]
+        #     }
+        # }
+    """
+    # Get metadata
+    metadata = get_tool_metadata(func)
+    
+    # Get function signature
+    sig = inspect.signature(func)
+    
+    # Try to get type hints
+    try:
+        type_hints = get_type_hints(func)
+    except Exception:
+        type_hints = {}
+    
+    # Build properties and required list
+    properties = {}
+    required = []
+    
+    for param_name, param in sig.parameters.items():
+        # Skip self, cls, and special parameters
+        if param_name in ('self', 'cls'):
+            continue
+        
+        # Skip *args and **kwargs
+        if param.kind in (inspect.Parameter.VAR_POSITIONAL, inspect.Parameter.VAR_KEYWORD):
+            continue
+        
+        # Get type annotation
+        param_type = type_hints.get(param_name, param.annotation)
+        
+        if param_type is inspect.Parameter.empty:
+            # No type annotation - default to string
+            properties[param_name] = {"type": "string"}
+        else:
+            properties[param_name] = python_type_to_json_schema(param_type)
+        
+        # Check if parameter is required (no default value)
+        if param.default is inspect.Parameter.empty:
+            # Check if it's Optional type
+            origin = get_origin(param_type) if param_type is not inspect.Parameter.empty else None
+            args = get_args(param_type) if param_type is not inspect.Parameter.empty else ()
+            
+            is_optional = origin is Union and type(None) in args
+            
+            if not is_optional:
+                required.append(param_name)
+    
+    # Build the schema
+    schema = {
+        "name": metadata["name"],
+        "description": metadata["description"],
+        "inputSchema": {
+            "type": "object",
+            "properties": properties,
+            "required": required
+        }
+    }
+    
+    return schema
+
+
+def functions_to_mcp_tools(functions: List[Callable]) -> List[Dict[str, Any]]:
+    """Convert multiple functions to MCP tool schemas.
+    
+    Args:
+        functions: List of functions to convert
+    
+    Returns:
+        List of MCP tool schemas
+    """
+    return [function_to_mcp_schema(func) for func in functions]

src/praisonai-agents/praisonaiagents/mcp/mcp_websocket.py

@@ -0,0 +1,596 @@
+"""
+WebSocket Transport for MCP (Model Context Protocol).
+
+This module provides WebSocket-based transport for connecting to MCP servers.
+WebSocket transport offers:
+- Long-lived bidirectional communication
+- Better cloud provider support (hibernatable connections)
+- Lower protocol overhead compared to HTTP
+- Session persistence across network interruptions
+
+This is an optional transport - the 'websockets' package is only imported
+when ws:// or wss:// URLs are used, ensuring no performance impact on
+existing stdio/SSE/HTTP transports.
+
+Based on SEP-1288: WebSocket Transport for MCP (in review).
+Protocol Revision: 2025-11-25
+"""
+
+import asyncio
+import json
+import logging
+import threading
+import inspect
+import re
+from typing import Any, Dict, Optional, List
+
+logger = logging.getLogger("mcp-websocket")
+
+
+def is_websocket_url(url: str) -> bool:
+    """
+    Check if a URL is a WebSocket URL.
+    
+    Args:
+        url: The URL string to check
+        
+    Returns:
+        True if the URL starts with ws:// or wss://, False otherwise
+    """
+    if not isinstance(url, str):
+        return False
+    return bool(re.match(r'^wss?://', url, re.IGNORECASE))
+
+
+def calculate_backoff(attempt: int, base_delay: float = 1.0, max_delay: float = 60.0) -> float:
+    """
+    Calculate exponential backoff delay for reconnection.
+    
+    Args:
+        attempt: The retry attempt number (0-indexed)
+        base_delay: Base delay in seconds
+        max_delay: Maximum delay cap in seconds
+        
+    Returns:
+        Delay in seconds for this attempt
+    """
+    delay = base_delay * (2 ** attempt)
+    return min(delay, max_delay)
+
+
+# Global event loop for WebSocket operations (similar to other MCP transports)
+_event_loop = None
+_loop_lock = threading.Lock()
+
+
+def get_event_loop():
+    """Get or create a global event loop for WebSocket operations."""
+    global _event_loop
+    with _loop_lock:
+        if _event_loop is None or _event_loop.is_closed():
+            _event_loop = asyncio.new_event_loop()
+        return _event_loop
+
+
+class WebSocketTransport:
+    """
+    Low-level WebSocket transport for MCP communication.
+    
+    This class handles the WebSocket connection and JSON-RPC message framing.
+    It supports:
+    - Secure (wss://) and insecure (ws://) connections
+    - Session ID management
+    - Authentication via subprotocol or headers
+    - Automatic reconnection with exponential backoff
+    """
+    
+    def __init__(
+        self,
+        url: str,
+        session_id: Optional[str] = None,
+        auth_token: Optional[str] = None,
+        max_retries: int = 3,
+        retry_delay: float = 1.0,
+        timeout: float = 30.0,
+        ping_interval: float = 30.0,
+        ping_timeout: float = 10.0
+    ):
+        """
+        Initialize WebSocket transport.
+        
+        Args:
+            url: WebSocket URL (ws:// or wss://)
+            session_id: Optional MCP session ID
+            auth_token: Optional authentication token
+            max_retries: Maximum reconnection attempts
+            retry_delay: Base delay between retries (exponential backoff)
+            timeout: Connection timeout in seconds
+            ping_interval: Interval between ping messages (keepalive)
+            ping_timeout: Timeout for ping response
+        """
+        self.url = url
+        self.session_id = session_id
+        self.auth_token = auth_token
+        self.max_retries = max_retries
+        self.retry_delay = retry_delay
+        self.timeout = timeout
+        self.ping_interval = ping_interval
+        self.ping_timeout = ping_timeout
+        
+        self._ws = None
+        self._closed = False
+        self._message_queue: asyncio.Queue = asyncio.Queue()
+        self._receive_task = None
+    
+    async def __aenter__(self):
+        """Async context manager entry - establish connection."""
+        await self.connect()
+        return self
+    
+    async def __aexit__(self, exc_type, exc_val, exc_tb):
+        """Async context manager exit - close connection."""
+        await self.close()
+    
+    async def connect(self):
+        """Establish WebSocket connection."""
+        try:
+            # Lazy import websockets
+            import websockets
+        except ImportError:
+            raise ImportError(
+                "websockets package is required for WebSocket transport. "
+                "Install it with: pip install websockets"
+            )
+        
+        # Build connection options
+        extra_headers = {}
+        subprotocols = []
+        
+        # Add session ID if present
+        if self.session_id:
+            extra_headers['Mcp-Session-Id'] = self.session_id
+        
+        # Add auth token via subprotocol (SEP-1288 recommendation)
+        if self.auth_token:
+            # Smuggle auth token in subprotocol list
+            # Format: "mcp-auth-<base64-encoded-token>"
+            import base64
+            encoded_token = base64.b64encode(self.auth_token.encode()).decode()
+            subprotocols.append(f"mcp-auth-{encoded_token}")
+        
+        # Always include MCP subprotocol
+        subprotocols.append("mcp")
+        
+        # Connect with retry logic
+        last_error = None
+        for attempt in range(self.max_retries + 1):
+            try:
+                self._ws = await asyncio.wait_for(
+                    websockets.connect(
+                        self.url,
+                        extra_headers=extra_headers if extra_headers else None,
+                        subprotocols=subprotocols if subprotocols else None,
+                        ping_interval=self.ping_interval,
+                        ping_timeout=self.ping_timeout,
+                        close_timeout=10
+                    ),
+                    timeout=self.timeout
+                )
+                logger.debug(f"WebSocket connected to {self.url}")
+                
+                # Start background receive task
+                self._receive_task = asyncio.create_task(self._receive_loop())
+                return
+                
+            except Exception as e:
+                last_error = e
+                if attempt < self.max_retries:
+                    delay = calculate_backoff(attempt, self.retry_delay)
+                    logger.warning(f"WebSocket connection failed, retrying in {delay}s: {e}")
+                    await asyncio.sleep(delay)
+                else:
+                    raise ConnectionError(f"Failed to connect after {self.max_retries + 1} attempts: {last_error}")
+    
+    async def _receive_loop(self):
+        """Background task to receive messages and queue them."""
+        try:
+            async for message in self._ws:
+                try:
+                    data = json.loads(message)
+                    await self._message_queue.put(data)
+                except json.JSONDecodeError as e:
+                    logger.error(f"Failed to parse WebSocket message: {e}")
+        except Exception as e:
+            if not self._closed:
+                logger.error(f"WebSocket receive error: {e}")
+                await self._message_queue.put({"error": str(e)})
+    
+    async def send(self, message: Dict[str, Any]):
+        """
+        Send a JSON-RPC message over WebSocket.
+        
+        Args:
+            message: JSON-RPC message dictionary
+        """
+        if self._ws is None:
+            raise RuntimeError("WebSocket not connected")
+        
+        data = json.dumps(message)
+        await self._ws.send(data)
+        logger.debug(f"Sent: {message.get('method', message.get('id', 'response'))}")
+    
+    async def receive(self) -> Dict[str, Any]:
+        """
+        Receive a JSON-RPC message from WebSocket.
+        
+        Returns:
+            Parsed JSON-RPC message dictionary
+        """
+        message = await self._message_queue.get()
+        if "error" in message and len(message) == 1:
+            raise ConnectionError(message["error"])
+        return message
+    
+    async def close(self):
+        """Close WebSocket connection."""
+        self._closed = True
+        
+        if self._receive_task:
+            self._receive_task.cancel()
+            try:
+                await self._receive_task
+            except asyncio.CancelledError:
+                pass
+        
+        if self._ws:
+            await self._ws.close()
+            self._ws = None
+        
+        logger.debug("WebSocket connection closed")
+
+
+class WebSocketMCPTool:
+    """
+    A wrapper for an MCP tool that can be used with praisonaiagents.
+    
+    This class wraps a tool from a WebSocket MCP server and makes it
+    compatible with the Agent class tool interface.
+    """
+    
+    def __init__(
+        self,
+        name: str,
+        description: str,
+        session: Any,  # ClientSession from mcp package
+        input_schema: Optional[Dict[str, Any]] = None,
+        timeout: int = 60
+    ):
+        """
+        Initialize WebSocket MCP tool wrapper.
+        
+        Args:
+            name: Tool name
+            description: Tool description
+            session: MCP ClientSession for making calls
+            input_schema: JSON Schema for tool input
+            timeout: Timeout for tool calls in seconds
+        """
+        self.name = name
+        self.__name__ = name
+        self.__qualname__ = name
+        self.__doc__ = description
+        self.description = description
+        self.session = session
+        self.input_schema = input_schema or {}
+        self.timeout = timeout
+        
+        # Build function signature from input schema
+        self.__signature__ = self._build_signature()
+    
+    def _build_signature(self) -> inspect.Signature:
+        """Build function signature from input schema."""
+        params = []
+        
+        if self.input_schema and 'properties' in self.input_schema:
+            required = self.input_schema.get('required', [])
+            
+            for param_name, prop_schema in self.input_schema['properties'].items():
+                # Determine type annotation
+                prop_type = prop_schema.get('type', 'string') if isinstance(prop_schema, dict) else 'string'
+                annotation = self._json_type_to_python(prop_type)
+                
+                # Determine default value
+                default = inspect.Parameter.empty if param_name in required else None
+                
+                params.append(
+                    inspect.Parameter(
+                        name=param_name,
+                        kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
+                        default=default,
+                        annotation=annotation
+                    )
+                )
+        
+        return inspect.Signature(params)
+    
+    def _json_type_to_python(self, json_type: str) -> type:
+        """Convert JSON Schema type to Python type."""
+        type_map = {
+            'string': str,
+            'integer': int,
+            'number': float,
+            'boolean': bool,
+            'array': list,
+            'object': dict
+        }
+        return type_map.get(json_type, Any)
+    
+    def __call__(self, **kwargs):
+        """
+        Call the tool synchronously.
+        
+        This wraps the async call for synchronous usage.
+        """
+        logger.debug(f"Tool {self.name} called with args: {kwargs}")
+        
+        loop = get_event_loop()
+        future = asyncio.run_coroutine_threadsafe(self._async_call(**kwargs), loop)
+        
+        try:
+            return future.result(timeout=self.timeout)
+        except Exception as e:
+            logger.error(f"Error calling tool {self.name}: {e}")
+            return f"Error: {str(e)}"
+    
+    async def _async_call(self, **kwargs):
+        """Call the tool asynchronously."""
+        logger.debug(f"Async calling tool {self.name} with args: {kwargs}")
+        
+        try:
+            result = await self.session.call_tool(self.name, kwargs)
+            
+            # Extract text from result
+            if hasattr(result, 'content') and result.content:
+                if hasattr(result.content[0], 'text'):
+                    return result.content[0].text
+                return str(result.content[0])
+            return str(result)
+        except Exception as e:
+            logger.error(f"Error in _async_call for {self.name}: {e}")
+            raise
+    
+    def _fix_array_schemas(self, schema: Dict[str, Any]) -> Dict[str, Any]:
+        """
+        Fix array schemas by adding missing 'items' attribute.
+        
+        OpenAI requires array types to specify item types.
+        """
+        if not isinstance(schema, dict):
+            return schema
+        
+        fixed_schema = schema.copy()
+        
+        # Fix array types at current level
+        if fixed_schema.get("type") == "array" and "items" not in fixed_schema:
+            fixed_schema["items"] = {"type": "string"}
+        
+        # Recursively fix nested schemas
+        if "properties" in fixed_schema:
+            fixed_properties = {}
+            for prop_name, prop_schema in fixed_schema["properties"].items():
+                fixed_properties[prop_name] = self._fix_array_schemas(prop_schema)
+            fixed_schema["properties"] = fixed_properties
+        
+        if "items" in fixed_schema:
+            fixed_schema["items"] = self._fix_array_schemas(fixed_schema["items"])
+        
+        return fixed_schema
+    
+    def to_openai_tool(self) -> Dict[str, Any]:
+        """Convert the tool to OpenAI function calling format."""
+        fixed_schema = self._fix_array_schemas(self.input_schema)
+        
+        return {
+            "type": "function",
+            "function": {
+                "name": self.name,
+                "description": self.description,
+                "parameters": fixed_schema
+            }
+        }
+
+
+class WebSocketMCPClient:
+    """
+    High-level client for connecting to MCP servers over WebSocket.
+    
+    This client handles:
+    - Connection establishment and lifecycle
+    - Tool discovery and wrapping
+    - Session management
+    - Automatic reconnection
+    
+    Example:
+        ```python
+        from praisonaiagents.mcp.mcp_websocket import WebSocketMCPClient
+        
+        client = WebSocketMCPClient("ws://localhost:8080/mcp")
+        for tool in client.tools:
+            print(tool.name)
+        ```
+    """
+    
+    def __init__(
+        self,
+        server_url: str,
+        debug: bool = False,
+        timeout: int = 60,
+        auth_token: Optional[str] = None,
+        options: Optional[Dict[str, Any]] = None
+    ):
+        """
+        Initialize WebSocket MCP client.
+        
+        Args:
+            server_url: WebSocket URL (ws:// or wss://)
+            debug: Enable debug logging
+            timeout: Timeout for operations in seconds
+            auth_token: Optional authentication token
+            options: Additional configuration options
+        """
+        self.server_url = server_url
+        self.debug = debug
+        self.timeout = timeout
+        self.auth_token = auth_token
+        self.options = options or {}
+        
+        self.session = None
+        self.transport = None
+        self.tools: List[WebSocketMCPTool] = []
+        self._closed = False
+        
+        # Configure logging
+        if debug:
+            logger.setLevel(logging.DEBUG)
+        else:
+            logger.setLevel(logging.WARNING)
+        
+        # Initialize connection
+        self._initialize()
+    
+    def _initialize(self):
+        """Initialize the connection and discover tools."""
+        loop = get_event_loop()
+        
+        # Start event loop in background thread if not running
+        def run_event_loop():
+            asyncio.set_event_loop(loop)
+            loop.run_forever()
+        
+        self.loop_thread = threading.Thread(target=run_event_loop, daemon=True)
+        self.loop_thread.start()
+        
+        # Run initialization
+        future = asyncio.run_coroutine_threadsafe(self._async_initialize(), loop)
+        self.tools = future.result(timeout=self.timeout)
+    
+    async def _async_initialize(self):
+        """Asynchronously initialize connection and discover tools."""
+        logger.debug(f"Connecting to MCP server at {self.server_url}")
+        
+        # Lazy import mcp package
+        try:
+            from mcp import ClientSession
+        except ImportError:
+            raise ImportError(
+                "mcp package is required. Install it with: pip install mcp"
+            )
+        
+        # Create transport
+        self.transport = WebSocketTransport(
+            self.server_url,
+            auth_token=self.auth_token,
+            timeout=self.timeout
+        )
+        await self.transport.connect()
+        
+        # Create read/write streams for ClientSession
+        # The ClientSession expects async iterators for read/write
+        read_stream = self._create_read_stream()
+        write_stream = self._create_write_stream()
+        
+        # Create and initialize session
+        self._session_context = ClientSession(read_stream, write_stream)
+        self.session = await self._session_context.__aenter__()
+        
+        await self.session.initialize()
+        
+        # Discover tools
+        logger.debug("Listing tools...")
+        response = await self.session.list_tools()
+        tools_data = response.tools
+        logger.debug(f"Found {len(tools_data)} tools: {[t.name for t in tools_data]}")
+        
+        # Create tool wrappers
+        tools = []
+        for tool in tools_data:
+            input_schema = tool.inputSchema if hasattr(tool, 'inputSchema') else None
+            wrapper = WebSocketMCPTool(
+                name=tool.name,
+                description=tool.description if hasattr(tool, 'description') else f"Call the {tool.name} tool",
+                session=self.session,
+                input_schema=input_schema,
+                timeout=self.timeout
+            )
+            tools.append(wrapper)
+        
+        return tools
+    
+    def _create_read_stream(self):
+        """Create async iterator for reading from transport."""
+        async def read():
+            while not self._closed:
+                try:
+                    message = await self.transport.receive()
+                    yield message
+                except Exception as e:
+                    if not self._closed:
+                        logger.error(f"Read error: {e}")
+                    break
+        return read()
+    
+    def _create_write_stream(self):
+        """Create write function for transport."""
+        async def write(message):
+            if hasattr(message, 'to_dict'):
+                message = message.to_dict()
+            await self.transport.send(message)
+        return write
+    
+    def __iter__(self):
+        """Return iterator over tools."""
+        return iter(self.tools)
+    
+    def to_openai_tools(self) -> List[Dict[str, Any]]:
+        """Convert all tools to OpenAI format."""
+        return [tool.to_openai_tool() for tool in self.tools]
+    
+    async def aclose(self):
+        """Async cleanup."""
+        if self._closed:
+            return
+        
+        self._closed = True
+        
+        try:
+            if hasattr(self, '_session_context') and self._session_context:
+                await self._session_context.__aexit__(None, None, None)
+        except Exception:
+            pass
+        
+        try:
+            if self.transport:
+                await self.transport.close()
+        except Exception:
+            pass
+    
+    def close(self):
+        """Synchronous cleanup."""
+        if self._closed:
+            return
+        
+        try:
+            loop = get_event_loop()
+            if not loop.is_closed():
+                future = asyncio.run_coroutine_threadsafe(self.aclose(), loop)
+                future.result(timeout=5)
+        except Exception:
+            self._closed = True
+    
+    def __del__(self):
+        """Cleanup on garbage collection."""
+        try:
+            if not self._closed:
+                self.close()
+        except Exception:
+            pass

src/praisonai-agents/praisonaiagents/memory/__init__.py

@@ -8,8 +8,149 @@ This module provides memory management capabilities including:
 - User memory for preferences/history
 - Quality-based storage decisions
 - Graph memory support via Mem0
+- Session save/resume (like Gemini CLI)
+- Context compression (like Gemini CLI)
+- Checkpointing (like Gemini CLI)
+- Rules management (like Cursor/Windsurf)
+- Auto-generated memories (like Windsurf Cascade)
+- Workflows (like Windsurf)
+- Hooks (like Windsurf Cascade Hooks)
+- Docs management (like Cursor docs)
+- MCP config management (like Cursor .cursor/mcp/)
+
+Memory Providers:
+- FileMemory: Zero-dependency JSON file-based storage (default)
+- Memory: Full-featured with SQLite, ChromaDB, Mem0, MongoDB support
+- RulesManager: Rules/instructions file management
+- AutoMemory: Automatic memory extraction from conversations
+- WorkflowManager: Multi-step workflow execution
+- HooksManager: Pre/post operation hooks
+- DocsManager: Documentation context management
+- MCPConfigManager: MCP server configuration management
 """
 
-from .memory import Memory
+from .file_memory import FileMemory, create_memory
+from .rules_manager import RulesManager, Rule, create_rules_manager
+from .docs_manager import DocsManager, Doc
+from .mcp_config import MCPConfigManager, MCPConfig
 
-__all__ = ["Memory"] 
+# Lazy imports for optional modules to avoid dependency issues and improve startup time
+def __getattr__(name):
+    if name == "Memory":
+        from .memory import Memory
+        return Memory
+    if name == "AutoMemory":
+        from .auto_memory import AutoMemory
+        return AutoMemory
+    if name == "AutoMemoryExtractor":
+        from .auto_memory import AutoMemoryExtractor
+        return AutoMemoryExtractor
+    if name == "create_auto_memory":
+        from .auto_memory import create_auto_memory
+        return create_auto_memory
+    # Workflows - now in dedicated module, re-export for backward compatibility
+    if name == "WorkflowManager":
+        from ..workflows import WorkflowManager
+        return WorkflowManager
+    if name == "Workflow":
+        from ..workflows import Workflow
+        return Workflow
+    if name == "WorkflowStep":
+        from ..workflows import WorkflowStep
+        return WorkflowStep
+    if name == "WorkflowContext":
+        from ..workflows import WorkflowContext
+        return WorkflowContext
+    if name == "StepResult":
+        from ..workflows import StepResult
+        return StepResult
+    if name == "Pipeline":
+        from ..workflows import Pipeline
+        return Pipeline
+    if name == "Route":
+        from ..workflows import Route
+        return Route
+    if name == "Parallel":
+        from ..workflows import Parallel
+        return Parallel
+    if name == "Loop":
+        from ..workflows import Loop
+        return Loop
+    if name == "Repeat":
+        from ..workflows import Repeat
+        return Repeat
+    if name == "route":
+        from ..workflows import route
+        return route
+    if name == "parallel":
+        from ..workflows import parallel
+        return parallel
+    if name == "loop":
+        from ..workflows import loop
+        return loop
+    if name == "repeat":
+        from ..workflows import repeat
+        return repeat
+    # Backward compatibility aliases
+    if name == "StepInput":
+        from ..workflows import WorkflowContext
+        return WorkflowContext
+    if name == "StepOutput":
+        from ..workflows import StepResult
+        return StepResult
+    if name == "create_workflow_manager":
+        from ..workflows.workflows import create_workflow_manager
+        return create_workflow_manager
+    if name == "HooksManager":
+        from .hooks import HooksManager
+        return HooksManager
+    if name == "HookResult":
+        from .hooks import HookResult
+        return HookResult
+    if name == "create_hooks_manager":
+        from .hooks import create_hooks_manager
+        return create_hooks_manager
+    # DocsManager and MCPConfigManager are already imported at module level
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+__all__ = [
+    # Core memory
+    "FileMemory", 
+    "Memory", 
+    "create_memory",
+    # Rules management
+    "RulesManager", 
+    "Rule", 
+    "create_rules_manager",
+    # Auto memory
+    "AutoMemory",
+    "AutoMemoryExtractor",
+    "create_auto_memory",
+    # Workflows (re-exported from praisonaiagents.workflows for backward compatibility)
+    "WorkflowManager",
+    "Workflow",
+    "Pipeline",
+    "WorkflowStep",
+    "WorkflowContext",
+    "StepResult",
+    "create_workflow_manager",
+    # Workflow patterns
+    "Route",
+    "Parallel",
+    "Loop",
+    "Repeat",
+    "route",
+    "parallel",
+    "loop",
+    "repeat",
+    # Hooks
+    "HooksManager",
+    "HookResult",
+    "create_hooks_manager",
+    # Docs management
+    "DocsManager",
+    "Doc",
+    # MCP config management
+    "MCPConfigManager",
+    "MCPConfig",
+] 

src/praisonai-agents/praisonaiagents/memory/auto_memory.py

@@ -0,0 +1,382 @@
+"""
+Auto-Generated Memory System for PraisonAI Agents.
+
+Provides automatic memory generation similar to Windsurf's Cascade,
+where the AI automatically decides what context is worth remembering.
+
+Features:
+- Automatic extraction of important facts from conversations
+- Entity recognition and storage
+- Preference detection
+- Lazy evaluation to avoid performance impact
+- Configurable importance thresholds
+
+This module is designed to be lightweight and only activates
+when explicitly enabled to avoid impacting performance.
+"""
+
+import re
+import logging
+from typing import Any, Dict, List, Optional, Callable, TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from .file_memory import FileMemory
+
+logger = logging.getLogger(__name__)
+
+
+class AutoMemoryExtractor:
+    """
+    Automatically extracts memorable content from conversations.
+    
+    Uses pattern matching for fast extraction without LLM calls,
+    with optional LLM enhancement for better accuracy.
+    
+    Example:
+        ```python
+        extractor = AutoMemoryExtractor()
+        
+        # Extract from conversation
+        memories = extractor.extract(
+            "My name is John and I prefer Python for backend work."
+        )
+        # Returns: [
+        #   {"type": "entity", "name": "John", "entity_type": "person"},
+        #   {"type": "preference", "content": "prefers Python for backend work"}
+        # ]
+        ```
+    """
+    
+    # Patterns for fast extraction (no LLM needed)
+    PATTERNS = {
+        # Name patterns
+        "name": [
+            r"(?:my name is|i'm|i am|call me)\s+([A-Z][a-z]+)",
+            r"(?:name|Name):\s*([A-Z][a-z]+)",
+        ],
+        # Preference patterns
+        "preference": [
+            r"i (?:prefer|like|love|enjoy|use)\s+(.+?)(?:\.|$)",
+            r"(?:my favorite|i always use)\s+(.+?)(?:\.|$)",
+        ],
+        # Role/job patterns
+        "role": [
+            r"i (?:am|work as|'m)\s+(?:a|an)\s+([a-z\s]+?)(?:\.|,|$)",
+            r"(?:my job|my role|i do)\s+(?:is|as)?\s*(.+?)(?:\.|$)",
+        ],
+        # Location patterns
+        "location": [
+            r"i (?:live|work|am based|am located)\s+(?:in|at)\s+([A-Z][a-z\s]+)",
+            r"(?:from|based in)\s+([A-Z][a-z\s,]+)",
+        ],
+        # Project/task patterns
+        "project": [
+            r"(?:working on|building|developing|creating)\s+(?:a|an|the)?\s*(.+?)(?:\.|$)",
+            r"(?:my project|the project)\s+(?:is|called)?\s*(.+?)(?:\.|$)",
+        ],
+        # Technology preferences
+        "technology": [
+            r"(?:using|use|prefer)\s+(Python|JavaScript|TypeScript|Go|Rust|Java|C\+\+|Ruby|PHP)",
+            r"(?:framework|library|tool)(?:s)?\s+(?:like|such as|including)\s+(.+?)(?:\.|$)",
+        ],
+    }
+    
+    # Importance scores by type
+    IMPORTANCE_SCORES = {
+        "name": 0.95,
+        "preference": 0.7,
+        "role": 0.85,
+        "location": 0.6,
+        "project": 0.75,
+        "technology": 0.7,
+    }
+    
+    def __init__(
+        self,
+        min_importance: float = 0.5,
+        use_llm: bool = False,
+        llm_func: Optional[Callable[[str], str]] = None,
+        verbose: int = 0
+    ):
+        """
+        Initialize AutoMemoryExtractor.
+        
+        Args:
+            min_importance: Minimum importance score to extract (0.0-1.0)
+            use_llm: Whether to use LLM for enhanced extraction
+            llm_func: LLM function for extraction (if use_llm=True)
+            verbose: Verbosity level
+        """
+        self.min_importance = min_importance
+        self.use_llm = use_llm
+        self.llm_func = llm_func
+        self.verbose = verbose
+        
+        # Compile patterns for performance
+        self._compiled_patterns = {}
+        for category, patterns in self.PATTERNS.items():
+            self._compiled_patterns[category] = [
+                re.compile(p, re.IGNORECASE) for p in patterns
+            ]
+    
+    def extract(self, text: str) -> List[Dict[str, Any]]:
+        """
+        Extract memorable content from text.
+        
+        Args:
+            text: Text to extract from (conversation, message, etc.)
+            
+        Returns:
+            List of extracted memories with type, content, and importance
+        """
+        memories = []
+        
+        # Fast pattern-based extraction
+        for category, patterns in self._compiled_patterns.items():
+            importance = self.IMPORTANCE_SCORES.get(category, 0.5)
+            
+            if importance < self.min_importance:
+                continue
+            
+            for pattern in patterns:
+                matches = pattern.findall(text)
+                for match in matches:
+                    if isinstance(match, tuple):
+                        match = match[0]
+                    
+                    match = match.strip()
+                    if len(match) < 2 or len(match) > 200:
+                        continue
+                    
+                    memory = {
+                        "type": category,
+                        "content": match,
+                        "importance": importance,
+                        "source": "pattern"
+                    }
+                    
+                    # Avoid duplicates
+                    if not any(m["content"].lower() == match.lower() for m in memories):
+                        memories.append(memory)
+        
+        # Optional LLM enhancement
+        if self.use_llm and self.llm_func and len(text) > 50:
+            llm_memories = self._extract_with_llm(text)
+            for mem in llm_memories:
+                if mem.get("importance", 0) >= self.min_importance:
+                    if not any(m["content"].lower() == mem["content"].lower() for m in memories):
+                        memories.append(mem)
+        
+        return memories
+    
+    def _extract_with_llm(self, text: str) -> List[Dict[str, Any]]:
+        """Extract memories using LLM for better accuracy."""
+        if not self.llm_func:
+            return []
+        
+        prompt = f"""Extract important facts from this text that should be remembered.
+Return a JSON array of objects with: type (name/preference/role/location/project/technology), content, importance (0.0-1.0).
+Only extract clear, specific facts. Be concise.
+
+Text: {text[:1500]}
+
+JSON array (no markdown, just the array):"""
+
+        try:
+            response = self.llm_func(prompt)
+            
+            # Parse JSON from response
+            import json
+            # Find JSON array in response
+            start = response.find("[")
+            end = response.rfind("]") + 1
+            if start >= 0 and end > start:
+                return json.loads(response[start:end])
+        except Exception as e:
+            if self.verbose:
+                logger.warning(f"LLM extraction failed: {e}")
+        
+        return []
+    
+    def should_remember(self, text: str) -> bool:
+        """
+        Quick check if text likely contains memorable content.
+        
+        Useful for filtering before full extraction.
+        
+        Args:
+            text: Text to check
+            
+        Returns:
+            True if text likely contains memorable content
+        """
+        # Quick keyword check
+        keywords = [
+            "my name", "i am", "i'm", "i prefer", "i like", "i use",
+            "i work", "i live", "working on", "building", "my project",
+            "favorite", "always use"
+        ]
+        
+        text_lower = text.lower()
+        return any(kw in text_lower for kw in keywords)
+
+
+class AutoMemory:
+    """
+    Wrapper that adds auto-generation capabilities to FileMemory.
+    
+    Intercepts interactions and automatically extracts/stores memories.
+    Designed to be lightweight - only processes when enabled.
+    
+    Example:
+        ```python
+        from praisonaiagents.memory import FileMemory
+        
+        base_memory = FileMemory(user_id="user123")
+        auto_memory = AutoMemory(base_memory, enabled=True)
+        
+        # Process conversation and auto-store memories
+        auto_memory.process_interaction(
+            user_message="My name is John and I prefer Python",
+            assistant_response="Nice to meet you, John!"
+        )
+        ```
+    """
+    
+    def __init__(
+        self,
+        memory: "FileMemory",
+        enabled: bool = True,
+        extractor: Optional[AutoMemoryExtractor] = None,
+        min_importance: float = 0.6,
+        verbose: int = 0
+    ):
+        """
+        Initialize AutoMemory wrapper.
+        
+        Args:
+            memory: Base FileMemory instance
+            enabled: Whether auto-generation is enabled
+            extractor: Custom extractor (or use default)
+            min_importance: Minimum importance to auto-store
+            verbose: Verbosity level
+        """
+        self.memory = memory
+        self.enabled = enabled
+        self.min_importance = min_importance
+        self.verbose = verbose
+        
+        self.extractor = extractor or AutoMemoryExtractor(
+            min_importance=min_importance,
+            verbose=verbose
+        )
+        
+        # Track what we've already processed to avoid duplicates
+        self._processed_hashes: set = set()
+    
+    def process_interaction(
+        self,
+        user_message: str,
+        assistant_response: Optional[str] = None,
+        store: bool = True
+    ) -> List[Dict[str, Any]]:
+        """
+        Process an interaction and extract memories.
+        
+        Args:
+            user_message: User's message
+            assistant_response: Optional assistant response
+            store: Whether to store extracted memories
+            
+        Returns:
+            List of extracted memories
+        """
+        if not self.enabled:
+            return []
+        
+        # Combine messages for extraction
+        text = user_message
+        if assistant_response:
+            text += "\n" + assistant_response
+        
+        # Check if already processed
+        import hashlib
+        text_hash = hashlib.md5(text.encode()).hexdigest()[:16]
+        if text_hash in self._processed_hashes:
+            return []
+        self._processed_hashes.add(text_hash)
+        
+        # Quick filter
+        if not self.extractor.should_remember(text):
+            return []
+        
+        # Extract memories
+        memories = self.extractor.extract(text)
+        
+        if store and memories:
+            self._store_memories(memories)
+        
+        return memories
+    
+    def _store_memories(self, memories: List[Dict[str, Any]]):
+        """Store extracted memories in the base memory."""
+        for mem in memories:
+            mem_type = mem.get("type", "")
+            content = mem.get("content", "")
+            importance = mem.get("importance", 0.5)
+            
+            if not content:
+                continue
+            
+            # Store based on type
+            if mem_type == "name":
+                # Store as entity
+                self.memory.add_entity(
+                    name=content,
+                    entity_type="person",
+                    attributes={"relationship": "user"}
+                )
+            elif mem_type in ("role", "location"):
+                # Store as entity attribute
+                self.memory.add_long_term(
+                    f"User's {mem_type}: {content}",
+                    importance=importance
+                )
+            else:
+                # Store as long-term memory
+                self.memory.add_long_term(content, importance=importance)
+            
+            if self.verbose:
+                logger.info(f"Auto-stored memory: {mem_type} - {content[:50]}...")
+    
+    def enable(self):
+        """Enable auto-generation."""
+        self.enabled = True
+    
+    def disable(self):
+        """Disable auto-generation."""
+        self.enabled = False
+    
+    def set_llm(self, llm_func: Callable[[str], str]):
+        """Set LLM function for enhanced extraction."""
+        self.extractor.use_llm = True
+        self.extractor.llm_func = llm_func
+
+
+def create_auto_memory(
+    memory: "FileMemory",
+    enabled: bool = True,
+    **kwargs
+) -> AutoMemory:
+    """
+    Create an AutoMemory wrapper.
+    
+    Args:
+        memory: Base FileMemory instance
+        enabled: Whether auto-generation is enabled
+        **kwargs: Additional configuration
+        
+    Returns:
+        AutoMemory instance
+    """
+    return AutoMemory(memory, enabled=enabled, **kwargs)

src/praisonai-agents/praisonaiagents/memory/docs_manager.py

@@ -0,0 +1,431 @@
+"""
+Docs Manager for PraisonAI Agents.
+
+Provides reusable documentation context support similar to:
+- Cursor Notepads (deprecated in favor of docs)
+- Cursor .cursor/docs/
+- Project documentation for AI context
+
+Features:
+- Auto-discovery of docs from .praison/docs/
+- Support for markdown, text, and code files
+- @doc mentions for including specific docs
+- Priority ordering for docs
+- Character limits for context window management
+
+Storage Structure:
+    .praison/docs/
+    ├── project-overview.md      # Project overview
+    ├── architecture.md          # Architecture decisions
+    ├── api-reference.md         # API documentation
+    └── coding-standards.md      # Coding guidelines
+
+Doc File Format (YAML frontmatter + Markdown):
+    ---
+    description: Project architecture overview
+    priority: 10
+    tags: [architecture, design]
+    ---
+    
+    # Architecture Overview
+    This project uses...
+"""
+
+import os
+import logging
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Literal
+from dataclasses import dataclass, field
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class Doc:
+    """A single documentation file with metadata and content."""
+    name: str
+    content: str
+    description: str = ""
+    priority: int = 0
+    tags: List[str] = field(default_factory=list)
+    file_path: Optional[str] = None
+    
+    def matches_tag(self, tag: str) -> bool:
+        """Check if this doc has a specific tag."""
+        return tag.lower() in [t.lower() for t in self.tags]
+    
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "name": self.name,
+            "content": self.content,
+            "description": self.description,
+            "priority": self.priority,
+            "tags": self.tags,
+            "file_path": self.file_path
+        }
+    
+    @classmethod
+    def from_dict(cls, data: Dict[str, Any]) -> 'Doc':
+        return cls(**data)
+
+
+class DocsManager:
+    """
+    Manages documentation context for AI agents.
+    
+    Provides:
+    - Auto-discovery of docs from .praison/docs/
+    - Global docs from ~/.praison/docs/
+    - @doc mentions for including specific docs
+    - Priority-based ordering
+    """
+    
+    DOCS_DIR_NAME = ".praison/docs"
+    SUPPORTED_EXTENSIONS = [".md", ".txt", ".rst"]
+    
+    # Maximum characters per doc (context window management)
+    MAX_DOC_CHARS = 12000
+    
+    def __init__(
+        self,
+        workspace_path: Optional[str] = None,
+        global_docs_path: Optional[str] = None,
+        verbose: int = 0
+    ):
+        """
+        Initialize DocsManager.
+        
+        Args:
+            workspace_path: Path to workspace/project root
+            global_docs_path: Path to global docs (default: ~/.praison/docs)
+            verbose: Verbosity level
+        """
+        self.workspace_path = Path(workspace_path) if workspace_path else Path.cwd()
+        self.global_docs_path = Path(global_docs_path) if global_docs_path else Path.home() / ".praison" / "docs"
+        self.verbose = verbose
+        
+        self._docs: Dict[str, Doc] = {}
+        self._load_all_docs()
+    
+    def _log(self, msg: str, level: int = logging.INFO):
+        """Log message if verbose."""
+        if self.verbose >= 1:
+            logger.log(level, msg)
+    
+    def _parse_frontmatter(self, content: str) -> tuple:
+        """
+        Parse YAML frontmatter from content.
+        
+        Returns:
+            Tuple of (frontmatter_dict, remaining_content)
+        """
+        frontmatter = {}
+        remaining = content
+        
+        # Check for YAML frontmatter
+        if content.startswith("---"):
+            parts = content.split("---", 2)
+            if len(parts) >= 3:
+                try:
+                    import yaml
+                    frontmatter = yaml.safe_load(parts[1]) or {}
+                    remaining = parts[2].strip()
+                except Exception:
+                    # If YAML parsing fails, use content as-is
+                    pass
+        
+        return frontmatter, remaining
+    
+    def _load_doc_file(self, file_path: Path, scope: str = "workspace") -> Optional[Doc]:
+        """Load a single doc file."""
+        try:
+            content = file_path.read_text(encoding="utf-8")
+            frontmatter, doc_content = self._parse_frontmatter(content)
+            
+            name = file_path.stem
+            
+            # Determine priority based on scope
+            base_priority = 100 if scope == "workspace" else 0
+            
+            doc = Doc(
+                name=name,
+                content=doc_content,
+                description=frontmatter.get("description", ""),
+                priority=frontmatter.get("priority", base_priority),
+                tags=frontmatter.get("tags", []),
+                file_path=str(file_path)
+            )
+            
+            return doc
+        except Exception as e:
+            self._log(f"Failed to load doc {file_path}: {e}", logging.WARNING)
+            return None
+    
+    def _load_docs_from_dir(self, docs_dir: Path, scope: str = "workspace"):
+        """Load all docs from a directory."""
+        if not docs_dir.exists():
+            return
+        
+        for ext in self.SUPPORTED_EXTENSIONS:
+            for file_path in docs_dir.glob(f"*{ext}"):
+                if file_path.is_file():
+                    doc = self._load_doc_file(file_path, scope)
+                    if doc:
+                        key = f"{scope}:{doc.name}"
+                        self._docs[key] = doc
+                        self._log(f"Loaded doc: {key}")
+    
+    def _load_all_docs(self):
+        """Load all docs from global and workspace directories."""
+        self._docs.clear()
+        
+        # 1. Load global docs (lowest priority)
+        if self.global_docs_path.exists():
+            self._load_docs_from_dir(self.global_docs_path, "global")
+            for key, doc in self._docs.items():
+                if key.startswith("global:"):
+                    doc.priority = doc.priority - 1000  # Lower priority for global
+        
+        # 2. Load workspace docs
+        workspace_docs_dir = self.workspace_path / self.DOCS_DIR_NAME.replace("/", os.sep)
+        if workspace_docs_dir.exists():
+            self._load_docs_from_dir(workspace_docs_dir, "workspace")
+        
+        self._log(f"Loaded {len(self._docs)} docs total")
+    
+    def reload(self):
+        """Reload all docs from disk."""
+        self._load_all_docs()
+    
+    def get_all_docs(self) -> List[Doc]:
+        """Get all loaded docs, sorted by priority."""
+        docs = list(self._docs.values())
+        docs.sort(key=lambda d: d.priority, reverse=True)
+        return docs
+    
+    def get_doc(self, name: str) -> Optional[Doc]:
+        """Get a specific doc by name."""
+        # Try workspace first, then global
+        for scope in ["workspace", "global"]:
+            key = f"{scope}:{name}"
+            if key in self._docs:
+                return self._docs[key]
+        return None
+    
+    def get_docs_by_tag(self, tag: str) -> List[Doc]:
+        """Get all docs with a specific tag."""
+        docs = [d for d in self._docs.values() if d.matches_tag(tag)]
+        docs.sort(key=lambda d: d.priority, reverse=True)
+        return docs
+    
+    def get_docs_for_context(
+        self,
+        include_docs: Optional[List[str]] = None,
+        include_tags: Optional[List[str]] = None,
+        max_chars: int = 50000
+    ) -> List[Doc]:
+        """
+        Get docs for inclusion in agent context.
+        
+        Args:
+            include_docs: Specific doc names to include (via @doc mentions)
+            include_tags: Tags to filter by
+            max_chars: Maximum total characters
+            
+        Returns:
+            List of docs sorted by priority
+        """
+        active = []
+        
+        # Include specific docs by name
+        if include_docs:
+            for name in include_docs:
+                doc = self.get_doc(name)
+                if doc and doc not in active:
+                    active.append(doc)
+        
+        # Include docs by tag
+        if include_tags:
+            for tag in include_tags:
+                for doc in self.get_docs_by_tag(tag):
+                    if doc not in active:
+                        active.append(doc)
+        
+        # If no specific docs requested, include all high-priority docs
+        if not include_docs and not include_tags:
+            for doc in self.get_all_docs():
+                if doc.priority >= 100:  # Only high-priority docs by default
+                    active.append(doc)
+        
+        # Sort by priority
+        active.sort(key=lambda d: d.priority, reverse=True)
+        
+        return active
+    
+    def format_docs_for_prompt(
+        self,
+        include_docs: Optional[List[str]] = None,
+        include_tags: Optional[List[str]] = None,
+        max_chars: int = 50000
+    ) -> str:
+        """
+        Format docs as a string for inclusion in prompts.
+        
+        Args:
+            include_docs: Specific doc names to include
+            include_tags: Tags to filter by
+            max_chars: Maximum total characters
+            
+        Returns:
+            Formatted docs string
+        """
+        active_docs = self.get_docs_for_context(include_docs, include_tags, max_chars)
+        
+        if not active_docs:
+            return ""
+        
+        parts = ["# Project Documentation\n"]
+        total_chars = len(parts[0])
+        
+        for doc in active_docs:
+            if total_chars >= max_chars:
+                break
+            
+            doc_text = doc.content.strip()
+            if not doc_text:
+                continue
+            
+            # Add doc with header
+            if doc.description:
+                header = f"## {doc.name}: {doc.description}"
+            else:
+                header = f"## {doc.name}"
+            
+            section = f"{header}\n{doc_text}\n"
+            
+            if total_chars + len(section) <= max_chars:
+                parts.append(section)
+                total_chars += len(section)
+            else:
+                # Truncate last doc
+                remaining = max_chars - total_chars
+                if remaining > 100:
+                    parts.append(section[:remaining] + "\n... (truncated)")
+                break
+        
+        return "\n".join(parts)
+    
+    def create_doc(
+        self,
+        name: str,
+        content: str,
+        description: str = "",
+        priority: int = 0,
+        tags: Optional[List[str]] = None,
+        scope: Literal["global", "workspace"] = "workspace"
+    ) -> Doc:
+        """
+        Create a new doc file.
+        
+        Args:
+            name: Doc name (used as filename)
+            content: Doc content (markdown)
+            description: Short description
+            priority: Priority (higher = applied first)
+            tags: Tags for categorization
+            scope: Where to save (global or workspace)
+            
+        Returns:
+            Created Doc object
+        """
+        # Determine save path
+        if scope == "global":
+            docs_dir = self.global_docs_path
+        else:
+            docs_dir = self.workspace_path / self.DOCS_DIR_NAME.replace("/", os.sep)
+        
+        docs_dir.mkdir(parents=True, exist_ok=True)
+        file_path = docs_dir / f"{name}.md"
+        
+        # Build frontmatter
+        frontmatter_lines = ["---"]
+        if description:
+            frontmatter_lines.append(f'description: "{description}"')
+        if priority != 0:
+            frontmatter_lines.append(f"priority: {priority}")
+        if tags:
+            tags_str = ", ".join(f'"{t}"' for t in tags)
+            frontmatter_lines.append(f"tags: [{tags_str}]")
+        frontmatter_lines.append("---")
+        frontmatter_lines.append("")
+        
+        # Write file
+        full_content = "\n".join(frontmatter_lines) + content
+        file_path.write_text(full_content, encoding="utf-8")
+        
+        # Create and register doc
+        doc = Doc(
+            name=name,
+            content=content,
+            description=description,
+            priority=priority,
+            tags=tags or [],
+            file_path=str(file_path)
+        )
+        
+        self._docs[f"{scope}:{name}"] = doc
+        self._log(f"Created doc '{name}' at {file_path}")
+        
+        return doc
+    
+    def delete_doc(self, name: str, scope: Optional[str] = None) -> bool:
+        """
+        Delete a doc file.
+        
+        Args:
+            name: Doc name to delete
+            scope: Scope to delete from (None = try both)
+            
+        Returns:
+            True if deleted, False if not found
+        """
+        scopes = [scope] if scope else ["workspace", "global"]
+        
+        for s in scopes:
+            key = f"{s}:{name}"
+            if key in self._docs:
+                doc = self._docs[key]
+                if doc.file_path:
+                    try:
+                        Path(doc.file_path).unlink()
+                        del self._docs[key]
+                        self._log(f"Deleted doc '{name}' from {s}")
+                        return True
+                    except Exception as e:
+                        self._log(f"Failed to delete doc file: {e}", logging.ERROR)
+        
+        return False
+    
+    def list_docs(self) -> List[Dict[str, Any]]:
+        """
+        List all docs with metadata.
+        
+        Returns:
+            List of doc info dicts
+        """
+        docs = []
+        for key, doc in self._docs.items():
+            scope = key.split(":")[0]
+            docs.append({
+                "name": doc.name,
+                "scope": scope,
+                "description": doc.description,
+                "priority": doc.priority,
+                "tags": doc.tags,
+                "file_path": doc.file_path,
+                "content_length": len(doc.content)
+            })
+        
+        # Sort by priority
+        docs.sort(key=lambda d: d["priority"], reverse=True)
+        return docs