Release v2.3.32

These tests make real API calls and should be excluded from CI.

.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/docker-publish.yml

@@ -1,45 +1,54 @@
-name: Daily Publish Docker
+name: Publish Docker
 
 on:
-  push:
-    branches:
-      - main
-  schedule:
-    - cron: '0 0 * * *'
+  release:
+    types: [published]
+  workflow_dispatch:
+    inputs:
+      tag:
+        description: 'Docker image tag (default: latest)'
+        required: false
+        default: 'latest'
 
 jobs:
-  check-and-build:
+  build-and-push:
     runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
     
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v2
-
-      - name: Check last commit time
-        id: check_commit
-        run: |
-          last_commit_time=$(git log -1 --format=%ct)
-          current_time=$(date +%s)
-          time_diff=$((current_time - last_commit_time))
-          if [ $time_diff -ge 120 ] && [ $time_diff -lt 240 ]; then
-            echo "should_run=true" >> $GITHUB_OUTPUT
-          else
-            echo "should_run=false" >> $GITHUB_OUTPUT
-          fi
+        uses: actions/checkout@v4
 
       - name: Set up Docker Buildx
-        if: steps.check_commit.outputs.should_run == 'true'
-        uses: docker/setup-buildx-action@v1
+        uses: docker/setup-buildx-action@v3
 
-      - name: Log in to the GitHub Container Registry
-        if: steps.check_commit.outputs.should_run == 'true'
-        run: echo ${{ secrets.GH_TOKEN }} | docker login ghcr.io -u USERNAME --password-stdin
+      - name: Log in to GitHub Container Registry
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Extract metadata for Docker
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ghcr.io/mervinpraison/praisonai
+          tags: |
+            type=raw,value=latest,enable=${{ github.event_name == 'release' }}
+            type=semver,pattern={{version}},enable=${{ github.event_name == 'release' }}
+            type=semver,pattern={{major}}.{{minor}},enable=${{ github.event_name == 'release' }}
+            type=raw,value=${{ github.event.inputs.tag || 'latest' }},enable=${{ github.event_name == 'workflow_dispatch' }}
 
       - name: Build and push Docker image
-        if: steps.check_commit.outputs.should_run == 'true'
-        uses: docker/build-push-action@v2
+        uses: docker/build-push-action@v5
         with:
           context: .
           file: ./docker/Dockerfile
           push: true
-          tags: ghcr.io/mervinpraison/praisonai:latest
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max

.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"

.github/workflows/test-comprehensive.yml

@@ -27,6 +27,9 @@ jobs:
   comprehensive-test:
     runs-on: ubuntu-latest
     timeout-minutes: 15
+    env:
+      OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+      OPENAI_MODEL_NAME: gpt-4o-mini
     strategy:
       matrix:
         python-version: ["3.11"]
@@ -55,11 +58,13 @@ jobs:
         # Install knowledge dependencies from praisonai-agents
         uv pip install --system "praisonaiagents[knowledge]"
 
-    - name: Set environment variables
+    - name: Verify environment variables
       run: |
-        echo "OPENAI_API_KEY=${{ secrets.OPENAI_API_KEY || 'sk-test-key-for-github-actions-testing-only-not-real' }}" >> $GITHUB_ENV
-        echo "OPENAI_API_BASE=${{ secrets.OPENAI_API_BASE || 'https://api.openai.com/v1' }}" >> $GITHUB_ENV
-        echo "OPENAI_MODEL_NAME=${{ secrets.OPENAI_MODEL_NAME || 'gpt-5-nano' }}" >> $GITHUB_ENV
+        if [ -n "$OPENAI_API_KEY" ]; then
+          echo "✅ OPENAI_API_KEY is set (length: ${#OPENAI_API_KEY})"
+        else
+          echo "⚠️ OPENAI_API_KEY is NOT set"
+        fi
 
     - name: Run Comprehensive Test Suite
       run: |

.github/workflows/test-core.yml

@@ -34,7 +34,7 @@ jobs:
         cd src/praisonai
         uv pip install --system ."[ui,gradio,api,agentops,google,openai,anthropic,cohere,chat,code,realtime,call,crewai,autogen]"
         uv pip install --system duckduckgo_search
-        uv pip install --system pytest pytest-asyncio pytest-cov
+        uv pip install --system pytest pytest-asyncio pytest-cov pytest-timeout
         # Install knowledge dependencies from praisonai-agents
         uv pip install --system "praisonaiagents[knowledge]"
 
@@ -262,7 +262,7 @@ jobs:
         export OPENAI_API_KEY="$OPENAI_API_KEY"
         export OPENAI_API_BASE="$OPENAI_API_BASE"
         export OPENAI_MODEL_NAME="$OPENAI_MODEL_NAME"
-        cd src/praisonai && python -m pytest tests/test.py -v --tb=short --disable-warnings
+        cd src/praisonai && python -m pytest tests/test.py -v --tb=short --disable-warnings -m "not real" --timeout=60
 
     - name: Upload coverage reports to Codecov
       if: matrix.python-version == '3.11'

.github/workflows/unittest.yml

@@ -27,7 +27,7 @@ jobs:
         cd src/praisonai
         uv pip install --system ."[ui,gradio,api,agentops,google,openai,anthropic,cohere,chat,code,realtime,call,crewai,autogen]"
         uv pip install --system duckduckgo_search
-        uv pip install --system pytest pytest-asyncio pytest-cov
+        uv pip install --system pytest pytest-asyncio pytest-cov pytest-timeout
         # Install knowledge dependencies from praisonai-agents
         uv pip install --system "praisonaiagents[knowledge]"
 
@@ -489,7 +489,7 @@ jobs:
 
     - name: Run Legacy Example Tests
       run: |
-        cd src/praisonai && python -m pytest tests/test.py -v --tb=short --disable-warnings
+        cd src/praisonai && python -m pytest tests/test.py -v --tb=short --disable-warnings -m "not real" --timeout=60
       continue-on-error: true
 
     - name: Upload coverage reports to Codecov

docker/Dockerfile

@@ -14,9 +14,10 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 RUN mkdir -p /root/.praison
 
 # Install Python packages (using latest versions)
+# Note: praisonai is installed from PyPI - remove version pin to use latest available
 RUN pip install --no-cache-dir \
     flask \
-    "praisonai>=2.2.91" \
+    praisonai \
     "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>=2.2.91" \
+    "praisonai>=2.3.32" \
     "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>=2.2.91" \
+    "praisonai>=2.3.32" \
     "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>=2.2.91" \
+    "praisonai>=2.3.32" \
     "praisonai[ui]" \
     "praisonai[crewai]"
 

docker/README.md

@@ -121,7 +121,7 @@ healthcheck:
 ## 📦 Package Versions
 
 All Docker images use consistent, up-to-date versions:
-- PraisonAI: `>=2.2.91`
+- PraisonAI: `>=2.2.95`
 - PraisonAI Agents: `>=0.0.92`
 - Python: `3.11-slim`
 
@@ -218,7 +218,7 @@ docker-compose up -d
 ### Version Pinning
 To use specific versions, update the Dockerfile:
 ```dockerfile
-RUN pip install "praisonai==2.2.91" "praisonaiagents==0.0.92"
+RUN pip install "praisonai==2.2.95" "praisonaiagents==0.0.92"
 ```
 
 ## 🌐 Production Deployment

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/cookbooks/yaml/aws_bedrock_agents.yaml

@@ -1,10 +1,10 @@
 framework: crewai
 topic: "Technology Research and Content Creation using AWS Bedrock"
 
-roles:
+agents:  # Canonical: use 'agents' instead of 'roles'
   research_specialist:
     role: "Research Specialist"
-    backstory: |
+    instructions:  # Canonical |
       You are an experienced technology researcher with deep expertise in analyzing 
       emerging technologies and market trends. You excel at gathering comprehensive 
       information from various sources and synthesizing it into actionable insights.
@@ -35,7 +35,7 @@ roles:
 
   content_creator:
     role: "Technical Content Creator"
-    backstory: |
+    instructions:  # Canonical |
       You are a skilled technical writer and content creator who specializes in 
       translating complex technology concepts into engaging, accessible content 
       for various audiences including developers, business leaders, and end users.
@@ -64,7 +64,7 @@ roles:
 
   quality_reviewer:
     role: "Content Quality Reviewer"
-    backstory: |
+    instructions:  # Canonical |
       You are a meticulous editor and quality assurance specialist with extensive 
       experience in technical content review. You ensure accuracy, clarity, and 
       consistency while maintaining the intended tone and message.

examples/cookbooks/yaml/secondary_market_research_agents.yaml

@@ -8,10 +8,10 @@ variables:
   industry: "Electric Vehicles"
   research_focus: "comprehensive market analysis"
 
-roles:
+agents:  # Canonical: use 'agents' instead of 'roles'
   market_overview_specialist:
     role: "Market Overview Specialist" 
-    backstory: |
+    instructions:  # Canonical |
       You are a seasoned market research analyst with 10+ years of experience 
       in analyzing market dynamics, trends, and growth patterns across various 
       industries. You excel at synthesizing complex market data into clear, 
@@ -64,7 +64,7 @@ roles:
 
   competitive_intelligence_analyst:
     role: "Competitive Intelligence Analyst"
-    backstory: |
+    instructions:  # Canonical |
       You are a competitive intelligence expert specializing in analyzing 
       competitive landscapes, market positioning, and strategic intelligence. 
       You have deep expertise in competitor analysis, market share assessment, 
@@ -119,7 +119,7 @@ roles:
 
   financial_performance_analyst:
     role: "Financial Performance Analyst"
-    backstory: |
+    instructions:  # Canonical |
       You are a financial analyst with expertise in corporate finance, 
       financial statement analysis, and industry benchmarking. You specialize 
       in analyzing financial performance metrics, trends, and comparative 
@@ -173,7 +173,7 @@ roles:
 
   growth_opportunities_researcher:
     role: "Growth Opportunities Researcher"
-    backstory: |
+    instructions:  # Canonical |
       You are a strategic business analyst specializing in identifying growth 
       opportunities, market expansion strategies, and emerging business models. 
       You excel at spotting trends, gaps, and untapped potential in markets 
@@ -228,7 +228,7 @@ roles:
 
   risk_assessment_specialist:
     role: "Risk Assessment Specialist"
-    backstory: |
+    instructions:  # Canonical |
       You are a risk management expert with extensive experience in identifying, 
       analyzing, and assessing business risks across various industries. You 
       specialize in market risks, competitive threats, regulatory challenges, 
@@ -287,7 +287,7 @@ roles:
 
   research_report_synthesizer:
     role: "Research Report Synthesizer"
-    backstory: |
+    instructions:  # Canonical |
       You are an expert business report writer and analyst with extensive 
       experience in synthesizing complex research findings into clear, 
       actionable business intelligence reports. You excel at creating 

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/multiple-mcp-servers.py

@@ -0,0 +1,108 @@
+"""
+Multiple MCP Servers Example
+
+This example demonstrates how to use multiple MCP servers with a single agent.
+You can combine different MCP servers (time, memory, filesystem, etc.) along
+with regular Python functions as tools.
+
+Requirements:
+- pip install praisonaiagents
+- uvx (for mcp-server-time)
+- npx (for @modelcontextprotocol servers)
+"""
+
+from praisonaiagents import Agent
+from praisonaiagents.mcp import MCP
+
+
+def my_custom_tool(query: str) -> str:
+    """
+    A custom Python function that can be used alongside MCP tools.
+    
+    Args:
+        query: The search query
+        
+    Returns:
+        A formatted result string
+    """
+    return f"Custom tool result for: {query}"
+
+
+def main():
+    # Example 1: Single MCP server (basic usage)
+    print("=" * 60)
+    print("Example 1: Single MCP Server")
+    print("=" * 60)
+    
+    agent_single = Agent(
+        name="TimeAgent",
+        instructions="You are a helpful assistant that can tell time.",
+        tools=MCP("uvx mcp-server-time")
+    )
+    
+    response = agent_single.start("What time is it in Tokyo?")
+    print(f"Response: {response}\n")
+    
+    # Example 2: Multiple MCP servers
+    print("=" * 60)
+    print("Example 2: Multiple MCP Servers")
+    print("=" * 60)
+    
+    agent_multi = Agent(
+        name="MultiMCPAgent",
+        instructions="""You are a helpful assistant with access to multiple tools:
+        - Time tools for getting current time and converting timezones
+        - Memory tools for storing and retrieving information
+        Use the appropriate tool based on the user's request.""",
+        tools=[
+            MCP("uvx mcp-server-time"),                     # Time tools
+            MCP("npx @modelcontextprotocol/server-memory"), # Memory tools
+        ]
+    )
+    
+    response = agent_multi.start("What time is it in New York and London?")
+    print(f"Response: {response}\n")
+    
+    # Example 3: Multiple MCP servers + custom functions
+    print("=" * 60)
+    print("Example 3: MCP Servers + Custom Functions")
+    print("=" * 60)
+    
+    agent_mixed = Agent(
+        name="MixedToolsAgent",
+        instructions="""You are a helpful assistant with access to:
+        - Time tools for timezone operations
+        - A custom search tool for general queries
+        Choose the right tool for each task.""",
+        tools=[
+            MCP("uvx mcp-server-time"),  # MCP server
+            my_custom_tool                # Regular Python function
+        ]
+    )
+    
+    response = agent_mixed.start("What time is it in Paris?")
+    print(f"Response: {response}\n")
+    
+    # Example 4: MCP with environment variables
+    print("=" * 60)
+    print("Example 4: MCP with Environment Variables")
+    print("=" * 60)
+    
+    # Note: Replace with your actual API key
+    # agent_with_env = Agent(
+    #     name="SearchAgent",
+    #     instructions="You can search the web for information.",
+    #     tools=[
+    #         MCP(
+    #             command="npx",
+    #             args=["-y", "@modelcontextprotocol/server-brave-search"],
+    #             env={"BRAVE_API_KEY": "your-api-key-here"}
+    #         ),
+    #         MCP("uvx mcp-server-time")
+    #     ]
+    # )
+    print("(Skipped - requires BRAVE_API_KEY)")
+
+
+if __name__ == "__main__":
+    main()

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/tools/e2b/agents.yaml

@@ -1,8 +1,8 @@
 framework: praisonai
 topic: crawl and extract all data from each url from this page https://news.ycombinator.com/
-roles:
+agents:  # Canonical: use 'agents' instead of 'roles'
   web_scraper:
-    backstory: An expert in data extraction from websites, adept at navigating and
+    instructions:  # Canonical An expert in data extraction from websites, adept at navigating and
       retrieving detailed information.
     goal: Extract URLs from the source page
     role: Web Scraper
@@ -15,7 +15,7 @@ roles:
     tools:
     - 'code_interpret'
   url_data_extractor:
-    backstory: Specializes in crawling websites to gather comprehensive data, ensuring
+    instructions:  # Canonical Specializes in crawling websites to gather comprehensive data, ensuring
       nothing is missed from each link.
     goal: Crawl each URL for data extraction
     role: URL Data Extractor
@@ -29,7 +29,7 @@ roles:
     tools:
     - 'code_interpret'
   data_organizer:
-    backstory: Known for keen attention to detail, capable of structuring vast amounts
+    instructions:  # Canonical Known for keen attention to detail, capable of structuring vast amounts
       of information into usable formats efficiently.
     goal: Organize extracted data into a structured format
     role: Data Organizer

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
+input: "Research AI trends and create a comprehensive report"  # The input passed to {{input}} in steps
+
+# 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
+agents:  # Canonical: use 'agents' instead of 'roles'
+  classifier:
+    role: Request Classifier
+    instructions:  # Canonical |
+      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
+    instructions:  # Canonical |
+      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
+    instructions:  # Canonical |
+      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
+    instructions:  # Canonical |
+      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
+    instructions:  # Canonical |
+      You analyze market trends and competitive landscapes.
+      You provide actionable business insights.
+    goal: Analyze market trends and opportunities
+    
+  aggregator:
+    role: Synthesizer
+    instructions:  # Canonical |
+      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 based on these results: {{previous_output}}"

examples/yaml/workflows/parallel_workflow.yaml

@@ -0,0 +1,47 @@
+# 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
+input: "AI-powered customer service solutions"  # The input passed to {{input}} in steps
+
+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: {{previous_output}}"

examples/yaml/workflows/routing_workflow.yaml

@@ -0,0 +1,50 @@
+# Routing Workflow
+# Demonstrates the route() pattern for decision-based branching
+
+name: Agentic Routing Workflow
+description: A classifier agent routes requests to specialized handler agents
+input: "Write a poem about the ocean"  # The input passed to {{input}} in steps
+
+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 +1,29 @@
+# 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
+input: "The benefits of renewable energy"  # The input passed to {{input}} in steps
+
+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 +86,5 @@ build
 graph.py
 chroma_db/
 .qodo
+*_COMPARISON_ANALYSIS.md
+*_REPORT.md

src/praisonai-agents/README.md

@@ -1,562 +0,0 @@
-# PraisonAI Agents
-
-A powerful Python framework for building AI agents with self-reflection, tool use, and multi-agent orchestration capabilities.
-
-[![PyPI version](https://badge.fury.io/py/praisonaiagents.svg)](https://badge.fury.io/py/praisonaiagents)
-[![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
-
-## Installation
-
-```bash
-# Basic installation
-pip install praisonaiagents
-
-# With all features
-pip install "praisonaiagents[all]"
-
-# Specific features
-pip install "praisonaiagents[memory]"      # Memory support
-pip install "praisonaiagents[knowledge]"   # Knowledge base
-pip install "praisonaiagents[mcp]"         # Model Context Protocol
-pip install "praisonaiagents[llm]"         # LiteLLM support
-pip install "praisonaiagents[mongodb]"     # MongoDB integration
-```
-
-## Quick Start
-
-### Basic Agent
-
-```python
-from praisonaiagents import Agent
-
-agent = Agent(
-    name="Assistant",
-    instructions="You are a helpful assistant.",
-    model="gpt-4o-mini"
-)
-
-response = agent.chat("What is the capital of France?")
-print(response)
-```
-
-### Agent with Tools
-
-```python
-from praisonaiagents import Agent
-
-def get_weather(city: str) -> str:
-    """Get weather for a city."""
-    return f"Weather in {city}: Sunny, 22°C"
-
-agent = Agent(
-    name="Weather Agent",
-    instructions="You help users check the weather.",
-    tools=[get_weather]
-)
-
-response = agent.chat("What's the weather in Paris?")
-```
-
-### Multi-Agent Workflow
-
-```python
-from praisonaiagents import Agent, Task, PraisonAIAgents
-
-# Define agents
-researcher = Agent(name="Researcher", instructions="Research topics thoroughly")
-writer = Agent(name="Writer", instructions="Write clear, engaging content")
-
-# Define tasks
-research_task = Task(
-    description="Research AI trends in 2025",
-    agent=researcher,
-    expected_output="Research findings"
-)
-
-write_task = Task(
-    description="Write an article based on research",
-    agent=writer,
-    expected_output="Article draft"
-)
-
-# Run workflow
-agents = PraisonAIAgents(agents=[researcher, writer], tasks=[research_task, write_task])
-result = agents.start()
-```
-
----
-
-## Agent Types
-
-### 1. Agent (Base)
-The core agent class with self-reflection, tool use, and streaming support.
-
-```python
-from praisonaiagents import Agent
-
-agent = Agent(
-    name="Assistant",
-    instructions="You are helpful.",
-    model="gpt-4o-mini",
-    verbose=True,
-    self_reflect=True,  # Enable self-reflection
-    max_reflect=3       # Max reflection iterations
-)
-```
-
-### 2. DeepResearchAgent 🆕
-Automated research agent using OpenAI or Gemini Deep Research APIs with real-time streaming.
-
-```python
-from praisonaiagents import DeepResearchAgent
-
-# OpenAI Deep Research
-agent = DeepResearchAgent(
-    model="o4-mini-deep-research",  # or "o3-deep-research"
-    verbose=True
-)
-
-result = agent.research("What are the latest AI trends?")
-print(result.report)
-print(f"Citations: {len(result.citations)}")
-
-# Gemini Deep Research
-agent = DeepResearchAgent(
-    model="deep-research-pro",  # Auto-detected as Gemini
-    verbose=True
-)
-
-result = agent.research("Research quantum computing advances")
-```
-
-**Features:**
-- **Multi-provider**: OpenAI, Gemini, and LiteLLM support
-- **Auto-detection**: Provider detected from model name
-- **Streaming**: Real-time progress with reasoning summaries (default: enabled)
-- **Citations**: Structured citations with URLs
-- **Tools**: Web search, code interpreter, MCP, file search
-
-### 3. ImageAgent
-Agent specialized for image generation and analysis.
-
-```python
-from praisonaiagents import ImageAgent
-
-agent = ImageAgent(
-    name="Artist",
-    instructions="Generate creative images"
-)
-
-result = agent.chat("Create an image of a sunset over mountains")
-```
-
-### 4. ContextAgent
-Agent with persistent context across conversations.
-
-```python
-from praisonaiagents import ContextAgent
-
-agent = ContextAgent(
-    name="Context Agent",
-    instructions="Remember our conversation"
-)
-```
-
-### 5. RouterAgent
-Routes requests to appropriate specialized agents.
-
-```python
-from praisonaiagents.agent import RouterAgent
-
-router = RouterAgent(
-    agents=[weather_agent, math_agent, general_agent],
-    instructions="Route to the best agent for each query"
-)
-```
-
----
-
-## Memory
-
-Persistent memory with short-term, long-term, and graph memory support.
-
-```python
-from praisonaiagents import Agent, Memory
-
-memory = Memory(
-    provider="chroma",  # or "mongodb", "rag"
-    use_short_term=True,
-    use_long_term=True
-)
-
-agent = Agent(
-    name="Memory Agent",
-    memory=memory
-)
-```
-
-**Memory Types:**
-- **Short-term**: Recent conversation context
-- **Long-term**: Persistent storage across sessions
-- **Graph**: Relationship-based memory with Mem0
-
----
-
-## Knowledge Base
-
-Add documents and data sources to your agents.
-
-```python
-from praisonaiagents import Agent, Knowledge
-
-knowledge = Knowledge(
-    sources=["docs/", "data.pdf", "https://example.com"],
-    chunking_method="semantic"
-)
-
-agent = Agent(
-    name="Knowledge Agent",
-    knowledge=knowledge
-)
-```
-
-**Supported Sources:**
-- PDF, Word, Excel, CSV files
-- Directories of documents
-- URLs and web pages
-- Databases
-
----
-
-## Tools
-
-### Built-in Tools
-
-```python
-from praisonaiagents.tools import (
-    # Search
-    duckduckgo_search,
-    searxng_search,
-    
-    # Data
-    read_csv, write_csv,
-    read_json, write_json,
-    read_yaml, write_yaml,
-    read_excel, write_excel,
-    
-    # Finance
-    get_stock_price, get_stock_info,
-    
-    # Research
-    arxiv_search,
-    wikipedia_search,
-    
-    # Web
-    spider_scrape,
-    newspaper_extract,
-    
-    # System
-    execute_shell,
-    run_python,
-    
-    # Database
-    duckdb_query,
-    mongodb_query
-)
-```
-
-### Custom Tools
-
-```python
-def my_tool(param: str) -> str:
-    """Tool description for the agent."""
-    return f"Result: {param}"
-
-agent = Agent(tools=[my_tool])
-```
-
----
-
-## MCP (Model Context Protocol)
-
-Connect to MCP servers for extended capabilities.
-
-```python
-from praisonaiagents import Agent, MCP
-
-# Local MCP server
-agent = Agent(
-    tools=MCP(
-        command="npx",
-        args=["-y", "@modelcontextprotocol/server-filesystem", "/path"]
-    )
-)
-
-# Remote MCP server (SSE)
-agent = Agent(
-    tools=MCP(url="http://localhost:8080/sse")
-)
-
-# With environment variables
-agent = Agent(
-    tools=MCP(
-        command="npx",
-        args=["-y", "@modelcontextprotocol/server-brave-search"],
-        env={"BRAVE_API_KEY": "your-key"}
-    )
-)
-```
-
----
-
-## Guardrails
-
-Add safety checks and validation to agent responses.
-
-```python
-from praisonaiagents import Agent, LLMGuardrail
-
-# LLM-based guardrail
-guardrail = LLMGuardrail(
-    instructions="Ensure responses are professional and accurate"
-)
-
-agent = Agent(
-    name="Safe Agent",
-    guardrail=guardrail
-)
-```
-
----
-
-## Handoffs
-
-Transfer conversations between agents.
-
-```python
-from praisonaiagents import Agent, handoff
-
-support_agent = Agent(name="Support", instructions="Handle support queries")
-sales_agent = Agent(name="Sales", instructions="Handle sales queries")
-
-# Enable handoffs
-support_agent.handoffs = [handoff(sales_agent, "Transfer to sales")]
-```
-
----
-
-## Streaming
-
-Real-time streaming responses.
-
-```python
-from praisonaiagents import Agent
-
-agent = Agent(name="Streamer", verbose=True)
-
-# Streaming is automatic with verbose=True
-response = agent.chat("Tell me a story")
-
-# Or use the stream method
-for chunk in agent.stream("Tell me a story"):
-    print(chunk, end="", flush=True)
-```
-
----
-
-## Async Support
-
-Full async/await support for all operations.
-
-```python
-import asyncio
-from praisonaiagents import Agent
-
-agent = Agent(name="Async Agent")
-
-async def main():
-    response = await agent.achat("Hello!")
-    print(response)
-
-asyncio.run(main())
-```
-
----
-
-## Telemetry
-
-Built-in telemetry for monitoring and analytics.
-
-```python
-from praisonaiagents import enable_telemetry, disable_telemetry
-
-# Enable full telemetry
-enable_telemetry()
-
-# Disable telemetry
-disable_telemetry()
-```
-
-**Environment Variables:**
-- `PRAISONAI_DISABLE_TELEMETRY=true` - Disable all telemetry
-- `PRAISONAI_PERFORMANCE_MODE=true` - Minimal overhead mode
-- `DO_NOT_TRACK=true` - Respect DNT preference
-
----
-
-## Supported LLM Providers
-
-PraisonAI Agents works with any LLM provider via LiteLLM:
-
-| Provider | Model Example |
-|----------|---------------|
-| OpenAI | `gpt-4o`, `gpt-4o-mini`, `o1`, `o3` |
-| Anthropic | `claude-3-5-sonnet`, `claude-3-opus` |
-| Google | `gemini-2.0-flash`, `gemini-pro` |
-| Ollama | `ollama/llama3`, `ollama/mistral` |
-| Azure | `azure/gpt-4` |
-| AWS Bedrock | `bedrock/claude-3` |
-| Groq | `groq/llama3-70b` |
-| Together | `together/mixtral-8x7b` |
-
-```python
-# Use any provider
-agent = Agent(model="anthropic/claude-3-5-sonnet")
-agent = Agent(model="ollama/llama3")
-agent = Agent(model="gemini/gemini-2.0-flash")
-```
-
----
-
-## Configuration
-
-### Environment Variables
-
-```bash
-# API Keys
-OPENAI_API_KEY=sk-...
-ANTHROPIC_API_KEY=sk-ant-...
-GEMINI_API_KEY=...
-GOOGLE_API_KEY=...
-
-# Logging
-LOGLEVEL=DEBUG  # DEBUG, INFO, WARNING, ERROR
-
-# Telemetry
-PRAISONAI_DISABLE_TELEMETRY=true
-```
-
----
-
-## Examples
-
-### Research Agent with Citations
-
-```python
-from praisonaiagents import DeepResearchAgent
-
-agent = DeepResearchAgent(
-    model="o4-mini-deep-research",
-    instructions="Provide data-rich insights with citations",
-    verbose=True
-)
-
-result = agent.research("Economic impact of AI on healthcare")
-
-print(result.report)
-for citation in result.citations:
-    print(f"- {citation.title}: {citation.url}")
-```
-
-### Multi-Agent Research Team
-
-```python
-from praisonaiagents import Agent, Task, PraisonAIAgents
-
-researcher = Agent(
-    name="Researcher",
-    instructions="Research topics thoroughly",
-    tools=[duckduckgo_search]
-)
-
-analyst = Agent(
-    name="Analyst", 
-    instructions="Analyze data and provide insights"
-)
-
-writer = Agent(
-    name="Writer",
-    instructions="Write clear reports"
-)
-
-tasks = [
-    Task(description="Research AI trends", agent=researcher),
-    Task(description="Analyze findings", agent=analyst),
-    Task(description="Write report", agent=writer)
-]
-
-team = PraisonAIAgents(agents=[researcher, analyst, writer], tasks=tasks)
-result = team.start()
-```
-
-### Agent with Memory and Knowledge
-
-```python
-from praisonaiagents import Agent, Memory, Knowledge
-
-agent = Agent(
-    name="Smart Assistant",
-    memory=Memory(use_long_term=True),
-    knowledge=Knowledge(sources=["docs/"]),
-    instructions="Help users with their questions using available knowledge"
-)
-```
-
----
-
-## API Reference
-
-### Agent
-
-| Parameter | Type | Default | Description |
-|-----------|------|---------|-------------|
-| `name` | str | "Agent" | Agent name |
-| `instructions` | str | "" | System instructions |
-| `model` | str | "gpt-4o-mini" | LLM model |
-| `tools` | list | [] | Available tools |
-| `memory` | Memory | None | Memory instance |
-| `knowledge` | Knowledge | None | Knowledge base |
-| `verbose` | bool | False | Enable verbose output |
-| `self_reflect` | bool | False | Enable self-reflection |
-| `max_reflect` | int | 3 | Max reflection iterations |
-
-### DeepResearchAgent
-
-| Parameter | Type | Default | Description |
-|-----------|------|---------|-------------|
-| `model` | str | "o3-deep-research" | Research model |
-| `provider` | str | None | Force provider (openai/gemini/litellm) |
-| `verbose` | bool | True | Show progress |
-| `poll_interval` | int | 5 | Gemini poll interval (seconds) |
-| `max_wait_time` | int | 3600 | Max research time (seconds) |
-
-### Task
-
-| Parameter | Type | Default | Description |
-|-----------|------|---------|-------------|
-| `description` | str | required | Task description |
-| `agent` | Agent | required | Assigned agent |
-| `expected_output` | str | "" | Expected output format |
-| `context` | list | [] | Context from other tasks |
-
----
-
-## License
-
-MIT License - see LICENSE file for details.
-
-## Links
-
-- [Documentation](https://docs.praison.ai)
-- [GitHub](https://github.com/MervinPraison/PraisonAI)
-- [PyPI](https://pypi.org/project/praisonaiagents/)

src/praisonai-agents/README.md

@@ -0,0 +1 @@
+../../README.md

src/praisonai-agents/benchmarks/deep_profiling.py

@@ -0,0 +1,368 @@
+#!/usr/bin/env python3
+"""
+PraisonAI Agents - Deep Profiling Benchmark
+
+Profiles each step of agent initialization to identify performance bottlenecks.
+
+Key Optimizations Applied:
+1. Lazy RulesManager - filesystem operations deferred until first access
+2. Lazy os.getcwd() - fast_context_path resolved only when needed
+3. Lazy console initialization - Rich Console created on demand
+
+Usage:
+    python benchmarks/deep_profiling.py
+"""
+
+import time
+import tracemalloc
+import cProfile
+import pstats
+import io
+from typing import Literal, Callable, Dict, List
+from dataclasses import dataclass
+
+# ============================================================================
+# Timing Utilities
+# ============================================================================
+
+@dataclass
+class StepTiming:
+    name: str
+    duration_us: float  # microseconds
+    memory_kb: float    # kilobytes
+
+
+class StepProfiler:
+    """Profile individual steps during initialization"""
+    
+    def __init__(self):
+        self.steps: List[StepTiming] = []
+        self._start_time = None
+        self._start_memory = None
+    
+    def start_step(self, name: str):
+        tracemalloc.start()
+        self._start_time = time.perf_counter()
+        self._step_name = name
+    
+    def end_step(self):
+        end_time = time.perf_counter()
+        current, peak = tracemalloc.get_traced_memory()
+        tracemalloc.stop()
+        
+        duration_us = (end_time - self._start_time) * 1_000_000
+        memory_kb = peak / 1024
+        
+        self.steps.append(StepTiming(
+            name=self._step_name,
+            duration_us=duration_us,
+            memory_kb=memory_kb
+        ))
+    
+    def print_report(self, title: str):
+        print(f"\n{'='*70}")
+        print(f"STEP-BY-STEP PROFILING: {title}")
+        print(f"{'='*70}")
+        
+        total_time = sum(s.duration_us for s in self.steps)
+        total_memory = sum(s.memory_kb for s in self.steps)
+        
+        print(f"\n{'Step':<40} {'Time (μs)':<15} {'Memory (KB)':<15} {'% Time':<10}")
+        print("-" * 80)
+        
+        for step in sorted(self.steps, key=lambda x: x.duration_us, reverse=True):
+            pct = (step.duration_us / total_time * 100) if total_time > 0 else 0
+            print(f"{step.name:<40} {step.duration_us:<15.2f} {step.memory_kb:<15.2f} {pct:<10.1f}%")
+        
+        print("-" * 80)
+        print(f"{'TOTAL':<40} {total_time:<15.2f} {total_memory:<15.2f} {'100.0':<10}%")
+
+
+def time_function(func: Callable, iterations: int = 100) -> Dict:
+    """Time a function with detailed statistics"""
+    times = []
+    
+    # Warmup
+    for _ in range(10):
+        func()
+    
+    # Measure
+    for _ in range(iterations):
+        start = time.perf_counter()
+        func()
+        end = time.perf_counter()
+        times.append((end - start) * 1_000_000)  # Convert to microseconds
+    
+    return {
+        'avg': sum(times) / len(times),
+        'min': min(times),
+        'max': max(times),
+        'median': sorted(times)[len(times)//2]
+    }
+
+
+# ============================================================================
+# Profile PraisonAI Agent Initialization - Step by Step
+# ============================================================================
+
+def profile_praisonai_detailed():
+    """Profile each step of PraisonAI Agent initialization"""
+    print("\n" + "="*70)
+    print("DETAILED PROFILING: PraisonAI Agent")
+    print("="*70)
+    
+    profiler = StepProfiler()
+    
+    # Step 1: Import
+    profiler.start_step("1. Import praisonaiagents")
+    from praisonaiagents import Agent as PraisonAgent
+    profiler.end_step()
+    
+    # Step 2: Define tool
+    profiler.start_step("2. Define tool function")
+    def get_weather(city: Literal["nyc", "sf"]):
+        """Use this to get weather information."""
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    tools = [get_weather]
+    profiler.end_step()
+    
+    # Step 3: Full instantiation
+    profiler.start_step("3. Full Agent instantiation")
+    agent = PraisonAgent(
+        name="Test Agent",
+        instructions="Be concise, reply with one sentence.",
+        llm="gpt-4o-mini",
+        tools=tools,
+        verbose=False
+    )
+    profiler.end_step()
+    
+    profiler.print_report("PraisonAI Agent")
+    
+    # Now profile the __init__ internals using cProfile
+    print("\n" + "="*70)
+    print("cProfile ANALYSIS: PraisonAI Agent.__init__")
+    print("="*70)
+    
+    pr = cProfile.Profile()
+    pr.enable()
+    
+    for _ in range(100):
+        agent = PraisonAgent(
+            name="Test Agent",
+            instructions="Be concise, reply with one sentence.",
+            llm="gpt-4o-mini",
+            tools=tools,
+            verbose=False
+        )
+    
+    pr.disable()
+    
+    # Print top 30 functions by cumulative time
+    s = io.StringIO()
+    ps = pstats.Stats(pr, stream=s).sort_stats('cumulative')
+    ps.print_stats(30)
+    print(s.getvalue())
+    
+    return agent
+
+
+# ============================================================================
+# Profile Agno Agent Initialization - Step by Step
+# ============================================================================
+
+def profile_agno_detailed():
+    """Profile each step of Agno Agent initialization"""
+    print("\n" + "="*70)
+    print("DETAILED PROFILING: Agno Agent")
+    print("="*70)
+    
+    profiler = StepProfiler()
+    
+    # Step 1: Import
+    profiler.start_step("1. Import agno")
+    from agno.agent import Agent as AgnoAgent
+    from agno.models.openai import OpenAIChat
+    profiler.end_step()
+    
+    # Step 2: Define tool
+    profiler.start_step("2. Define tool function")
+    def get_weather(city: Literal["nyc", "sf"]):
+        """Use this to get weather information."""
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    tools = [get_weather]
+    profiler.end_step()
+    
+    # Step 3: Full instantiation
+    profiler.start_step("3. Full Agent instantiation")
+    agent = AgnoAgent(model=OpenAIChat(id="gpt-4o-mini"), tools=tools)
+    profiler.end_step()
+    
+    profiler.print_report("Agno Agent")
+    
+    # Now profile the __init__ internals using cProfile
+    print("\n" + "="*70)
+    print("cProfile ANALYSIS: Agno Agent.__init__")
+    print("="*70)
+    
+    pr = cProfile.Profile()
+    pr.enable()
+    
+    for _ in range(100):
+        agent = AgnoAgent(model=OpenAIChat(id="gpt-4o-mini"), tools=tools)
+    
+    pr.disable()
+    
+    # Print top 30 functions by cumulative time
+    s = io.StringIO()
+    ps = pstats.Stats(pr, stream=s).sort_stats('cumulative')
+    ps.print_stats(30)
+    print(s.getvalue())
+    
+    return agent
+
+
+# ============================================================================
+# Profile Individual Components of PraisonAI
+# ============================================================================
+
+def profile_praisonai_components():
+    """Profile individual components of PraisonAI Agent initialization"""
+    print("\n" + "="*70)
+    print("COMPONENT-LEVEL PROFILING: PraisonAI Agent")
+    print("="*70)
+    
+    from praisonaiagents import Agent as PraisonAgent
+    
+    def get_weather(city: Literal["nyc", "sf"]):
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    tools = [get_weather]
+    
+    # Test different configurations
+    configs = [
+        ("Minimal (no tools, no llm)", {"name": "Test", "verbose": False}),
+        ("With name only", {"name": "Test Agent", "verbose": False}),
+        ("With instructions", {"name": "Test", "instructions": "Be concise", "verbose": False}),
+        ("With llm (gpt-4o-mini)", {"name": "Test", "llm": "gpt-4o-mini", "verbose": False}),
+        ("With llm (openai/gpt-4o-mini)", {"name": "Test", "llm": "openai/gpt-4o-mini", "verbose": False}),
+        ("With tools", {"name": "Test", "tools": tools, "verbose": False}),
+        ("Full config (OpenAI SDK)", {"name": "Test", "instructions": "Be concise", "llm": "gpt-4o-mini", "tools": tools, "verbose": False}),
+        ("Full config (LiteLLM)", {"name": "Test", "instructions": "Be concise", "llm": "openai/gpt-4o-mini", "tools": tools, "verbose": False}),
+    ]
+    
+    print(f"\n{'Configuration':<40} {'Avg Time (μs)':<15} {'Min (μs)':<15} {'Max (μs)':<15}")
+    print("-" * 85)
+    
+    for name, config in configs:
+        def create_agent():
+            return PraisonAgent(**config)
+        
+        stats = time_function(create_agent, iterations=100)
+        print(f"{name:<40} {stats['avg']:<15.2f} {stats['min']:<15.2f} {stats['max']:<15.2f}")
+
+
+# ============================================================================
+# Compare Specific Operations
+# ============================================================================
+
+def compare_specific_operations():
+    """Compare specific operations between PraisonAI and Agno"""
+    print("\n" + "="*70)
+    print("OPERATION COMPARISON: PraisonAI vs Agno")
+    print("="*70)
+    
+    from praisonaiagents import Agent as PraisonAgent
+    from agno.agent import Agent as AgnoAgent
+    from agno.models.openai import OpenAIChat
+    
+    def get_weather(city: Literal["nyc", "sf"]):
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    tools = [get_weather]
+    
+    operations = []
+    
+    # Operation 1: Bare minimum instantiation
+    def praisonai_minimal():
+        return PraisonAgent(name="Test", verbose=False)
+    
+    def agno_minimal():
+        return AgnoAgent(model=OpenAIChat(id="gpt-4o-mini"))
+    
+    praison_stats = time_function(praisonai_minimal, 100)
+    agno_stats = time_function(agno_minimal, 100)
+    operations.append(("Minimal instantiation", praison_stats['avg'], agno_stats['avg']))
+    
+    # Operation 2: With tools
+    def praisonai_with_tools():
+        return PraisonAgent(name="Test", tools=tools, verbose=False)
+    
+    def agno_with_tools():
+        return AgnoAgent(model=OpenAIChat(id="gpt-4o-mini"), tools=tools)
+    
+    praison_stats = time_function(praisonai_with_tools, 100)
+    agno_stats = time_function(agno_with_tools, 100)
+    operations.append(("With tools", praison_stats['avg'], agno_stats['avg']))
+    
+    # Operation 3: Full config
+    def praisonai_full():
+        return PraisonAgent(
+            name="Test Agent",
+            instructions="Be concise",
+            llm="gpt-4o-mini",
+            tools=tools,
+            verbose=False
+        )
+    
+    def agno_full():
+        return AgnoAgent(
+            model=OpenAIChat(id="gpt-4o-mini"),
+            tools=tools,
+            instructions=["Be concise"]
+        )
+    
+    praison_stats = time_function(praisonai_full, 100)
+    agno_stats = time_function(agno_full, 100)
+    operations.append(("Full config", praison_stats['avg'], agno_stats['avg']))
+    
+    print(f"\n{'Operation':<30} {'PraisonAI (μs)':<18} {'Agno (μs)':<18} {'Ratio':<10}")
+    print("-" * 76)
+    
+    for op_name, praison_time, agno_time in operations:
+        ratio = praison_time / agno_time if agno_time > 0 else 0
+        print(f"{op_name:<30} {praison_time:<18.2f} {agno_time:<18.2f} {ratio:<10.1f}x")
+
+
+# ============================================================================
+# Main
+# ============================================================================
+
+if __name__ == "__main__":
+    import os
+    os.environ.setdefault("OPENAI_API_KEY", "sk-test")
+    
+    print("="*70)
+    print("PraisonAI Agents - Deep Profiling Benchmark")
+    print("="*70)
+    print("\nThis analysis identifies performance bottlenecks in agent initialization.\n")
+    
+    # Run all profiling
+    profile_praisonai_detailed()
+    profile_agno_detailed()
+    profile_praisonai_components()
+    compare_specific_operations()
+    
+    print("\n" + "="*70)
+    print("PROFILING COMPLETE")
+    print("="*70)

src/praisonai-agents/benchmarks/performance_benchmark.py

@@ -0,0 +1,416 @@
+"""
+PraisonAI Agents - Comprehensive Performance Benchmark
+
+Measures agent instantiation time and memory footprint across
+multiple AI agent frameworks using industry-standard methodology.
+
+Usage:
+    python benchmarks/performance_benchmark.py
+"""
+
+import gc
+import time
+import tracemalloc
+import statistics
+from typing import Literal, List, Callable
+from dataclasses import dataclass, field
+
+
+@dataclass
+class PerformanceResult:
+    """Performance measurement results"""
+    run_times: List[float] = field(default_factory=list)
+    memory_usages: List[float] = field(default_factory=list)
+    
+    @property
+    def avg_run_time(self) -> float:
+        return statistics.mean(self.run_times) if self.run_times else 0
+    
+    @property
+    def avg_memory_usage(self) -> float:
+        return statistics.mean(self.memory_usages) if self.memory_usages else 0
+    
+    @property
+    def min_run_time(self) -> float:
+        return min(self.run_times) if self.run_times else 0
+    
+    @property
+    def max_run_time(self) -> float:
+        return max(self.run_times) if self.run_times else 0
+    
+    @property
+    def std_dev_run_time(self) -> float:
+        return statistics.stdev(self.run_times) if len(self.run_times) > 1 else 0
+    
+    @property
+    def median_run_time(self) -> float:
+        return statistics.median(self.run_times) if self.run_times else 0
+
+
+def compute_tracemalloc_baseline(samples: int = 3) -> float:
+    """Compute baseline memory usage for tracemalloc"""
+    def empty_func():
+        return
+    
+    results = []
+    for _ in range(samples):
+        gc.collect()
+        tracemalloc.start()
+        empty_func()
+        _, peak = tracemalloc.get_traced_memory()
+        tracemalloc.stop()
+        results.append(peak / 1024 / 1024)  # Convert to MiB
+    
+    return sum(results) / len(results) if results else 0
+
+
+def measure_time(func: Callable, iterations: int = 1000, warmup: int = 10) -> List[float]:
+    """Measure execution time for a function"""
+    # Warmup runs
+    for _ in range(warmup):
+        func()
+    
+    # Measured runs
+    times = []
+    for _ in range(iterations):
+        start = time.perf_counter()
+        func()
+        end = time.perf_counter()
+        times.append(end - start)
+    
+    return times
+
+
+def measure_memory(func: Callable, iterations: int = 1000, warmup: int = 10) -> List[float]:
+    """Measure memory usage for a function"""
+    baseline = compute_tracemalloc_baseline()
+    
+    # Warmup runs
+    for _ in range(warmup):
+        func()
+    
+    # Measured runs
+    memory_usages = []
+    for _ in range(iterations):
+        gc.collect()
+        tracemalloc.start()
+        func()
+        _, peak = tracemalloc.get_traced_memory()
+        tracemalloc.stop()
+        
+        peak_mib = peak / 1024 / 1024
+        adjusted = max(0, peak_mib - baseline)
+        memory_usages.append(adjusted)
+    
+    return memory_usages
+
+
+def run_benchmark(func: Callable, name: str, iterations: int = 1000, warmup: int = 10) -> PerformanceResult:
+    """Run full benchmark for a function"""
+    print(f"\n{'='*60}")
+    print(f"Benchmarking: {name}")
+    print(f"{'='*60}")
+    print(f"Warmup runs: {warmup}")
+    print(f"Measured iterations: {iterations}")
+    
+    print("\nMeasuring runtime...")
+    run_times = measure_time(func, iterations, warmup)
+    
+    print("Measuring memory...")
+    memory_usages = measure_memory(func, iterations, warmup)
+    
+    result = PerformanceResult(run_times=run_times, memory_usages=memory_usages)
+    
+    print(f"\nResults for {name}:")
+    print(f"  Average Time: {result.avg_run_time:.6f} seconds ({result.avg_run_time * 1_000_000:.2f} μs)")
+    print(f"  Min Time:     {result.min_run_time:.6f} seconds")
+    print(f"  Max Time:     {result.max_run_time:.6f} seconds")
+    print(f"  Std Dev:      {result.std_dev_run_time:.6f} seconds")
+    print(f"  Median Time:  {result.median_run_time:.6f} seconds")
+    print(f"  Avg Memory:   {result.avg_memory_usage:.6f} MiB ({result.avg_memory_usage * 1024:.2f} KiB)")
+    
+    return result
+
+
+# ============================================================================
+# Framework Imports
+# ============================================================================
+from praisonaiagents import Agent as PraisonAgent
+from agno.agent import Agent as AgnoAgent
+from agno.models.openai import OpenAIChat
+
+# Additional frameworks for comparison
+try:
+    from pydantic_ai import Agent as PydanticAgent
+    PYDANTIC_AVAILABLE = True
+except ImportError:
+    PYDANTIC_AVAILABLE = False
+
+try:
+    from crewai.agent import Agent as CrewAgent
+    from crewai.tools import tool as crewai_tool
+    CREWAI_AVAILABLE = True
+except ImportError:
+    CREWAI_AVAILABLE = False
+
+try:
+    from langchain_core.tools import tool as langchain_tool
+    from langchain_openai import ChatOpenAI
+    from langgraph.prebuilt import create_react_agent
+    LANGGRAPH_AVAILABLE = True
+except ImportError:
+    LANGGRAPH_AVAILABLE = False
+
+try:
+    from agents import Agent as OpenAIAgent, function_tool
+    OPENAI_AGENTS_AVAILABLE = True
+except ImportError:
+    OPENAI_AGENTS_AVAILABLE = False
+
+
+def get_weather(city: Literal["nyc", "sf"]):
+    """Use this to get weather information."""
+    if city == "nyc":
+        return "It might be cloudy in nyc"
+    elif city == "sf":
+        return "It's always sunny in sf"
+
+
+tools = [get_weather]
+
+
+def instantiate_praisonai_agent():
+    """Instantiate a PraisonAI agent with one tool."""
+    return PraisonAgent(
+        name="Test Agent",
+        instructions="Be concise, reply with one sentence.",
+        llm="gpt-4o-mini",
+        tools=tools,
+        verbose=False
+    )
+
+
+def instantiate_praisonai_litellm_agent():
+    """Instantiate a PraisonAI agent with LiteLLM backend."""
+    return PraisonAgent(
+        name="Test Agent",
+        instructions="Be concise, reply with one sentence.",
+        llm="openai/gpt-4o-mini",
+        tools=tools,
+        verbose=False
+    )
+
+
+def instantiate_agno_agent():
+    """Instantiate an Agno agent with one tool."""
+    return AgnoAgent(model=OpenAIChat(id="gpt-4o-mini"), tools=tools)
+
+
+def instantiate_pydantic_agent():
+    """Instantiate a PydanticAI agent with one tool."""
+    agent = PydanticAgent("openai:gpt-4o-mini", system_prompt="Be concise, reply with one sentence.")
+    @agent.tool_plain
+    def get_weather_pydantic(city: Literal["nyc", "sf"]):
+        """Use this to get weather information."""
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    return agent
+
+
+def instantiate_openai_agents():
+    """Instantiate an OpenAI Agents SDK agent with one tool."""
+    return OpenAIAgent(
+        name="Test agent",
+        instructions="Be concise, reply with one sentence.",
+        model="gpt-4o-mini",
+        tools=[function_tool(get_weather)],
+    )
+
+
+def instantiate_langgraph_agent():
+    """Instantiate a LangGraph agent with one tool."""
+    @langchain_tool
+    def get_weather_lg(city: Literal["nyc", "sf"]):
+        """Use this to get weather information."""
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    return create_react_agent(model=ChatOpenAI(model="gpt-4o-mini"), tools=[get_weather_lg])
+
+
+def instantiate_crewai_agent():
+    """Instantiate a CrewAI agent with one tool."""
+    @crewai_tool("Weather Tool")
+    def get_weather_crew(city: Literal["nyc", "sf"]):
+        """Use this to get weather information."""
+        if city == "nyc":
+            return "It might be cloudy in nyc"
+        elif city == "sf":
+            return "It's always sunny in sf"
+    return CrewAgent(
+        llm="gpt-4o-mini",
+        role="Test Agent",
+        goal="Be concise, reply with one sentence.",
+        tools=[get_weather_crew],
+        backstory="Test",
+    )
+
+
+if __name__ == "__main__":
+    print("="*70)
+    print("PraisonAI Agents - Comprehensive Performance Benchmark")
+    print("="*70)
+    print("\nThis benchmark measures:")
+    print("  1. Agent instantiation time (how fast an agent object is created)")
+    print("  2. Memory footprint (how much memory an agent uses)")
+    print("\nMethodology:")
+    print("  - 10 warmup runs (not measured)")
+    print("  - 1000 measured iterations")
+    print("  - tracemalloc for memory measurement")
+    print("  - Baseline subtraction for accuracy")
+    print("\nFrameworks available:")
+    print("  - PraisonAI: ✅")
+    print(f"  - Agno: ✅")
+    print(f"  - PydanticAI: {'✅' if PYDANTIC_AVAILABLE else '❌'}")
+    print(f"  - OpenAI Agents SDK: {'✅' if OPENAI_AGENTS_AVAILABLE else '❌'}")
+    print(f"  - LangGraph: {'✅' if LANGGRAPH_AVAILABLE else '❌'}")
+    print(f"  - CrewAI: {'✅' if CREWAI_AVAILABLE else '❌'}")
+    
+    results = {}
+    
+    # Test PraisonAI first
+    try:
+        print("\n" + "="*70)
+        print("Testing PraisonAI Agents")
+        print("="*70)
+        results['praisonai'] = run_benchmark(
+            instantiate_praisonai_agent,
+            "PraisonAI",
+            iterations=1000,
+            warmup=10
+        )
+    except Exception as e:
+        print(f"Error testing PraisonAI: {e}")
+    
+    try:
+        print("\n" + "="*70)
+        print("Testing PraisonAI Agents (LiteLLM)")
+        print("="*70)
+        results['praisonai_litellm'] = run_benchmark(
+            instantiate_praisonai_litellm_agent,
+            "PraisonAI (LiteLLM)",
+            iterations=1000,
+            warmup=10
+        )
+    except Exception as e:
+        print(f"Error testing PraisonAI (LiteLLM): {e}")
+    
+    # Test Agno
+    try:
+        print("\n" + "="*70)
+        print("Testing Agno")
+        print("="*70)
+        results['agno'] = run_benchmark(
+            instantiate_agno_agent,
+            "Agno",
+            iterations=1000,
+            warmup=10
+        )
+    except Exception as e:
+        print(f"Error testing Agno: {e}")
+    
+    # Test PydanticAI
+    if PYDANTIC_AVAILABLE:
+        try:
+            print("\n" + "="*70)
+            print("Testing PydanticAI")
+            print("="*70)
+            results['pydantic'] = run_benchmark(
+                instantiate_pydantic_agent,
+                "PydanticAI Agent Instantiation",
+                iterations=1000,
+                warmup=10
+            )
+        except Exception as e:
+            print(f"Error testing PydanticAI: {e}")
+    
+    # Test OpenAI Agents SDK
+    if OPENAI_AGENTS_AVAILABLE:
+        try:
+            print("\n" + "="*70)
+            print("Testing OpenAI Agents SDK")
+            print("="*70)
+            results['openai_agents'] = run_benchmark(
+                instantiate_openai_agents,
+                "OpenAI Agents SDK Instantiation",
+                iterations=1000,
+                warmup=10
+            )
+        except Exception as e:
+            print(f"Error testing OpenAI Agents SDK: {e}")
+    
+    # Test LangGraph
+    if LANGGRAPH_AVAILABLE:
+        try:
+            print("\n" + "="*70)
+            print("Testing LangGraph")
+            print("="*70)
+            results['langgraph'] = run_benchmark(
+                instantiate_langgraph_agent,
+                "LangGraph Agent Instantiation",
+                iterations=1000,
+                warmup=10
+            )
+        except Exception as e:
+            print(f"Error testing LangGraph: {e}")
+    
+    # Test CrewAI
+    if CREWAI_AVAILABLE:
+        try:
+            print("\n" + "="*70)
+            print("Testing CrewAI")
+            print("="*70)
+            results['crewai'] = run_benchmark(
+                instantiate_crewai_agent,
+                "CrewAI Agent Instantiation",
+                iterations=1000,
+                warmup=10
+            )
+        except Exception as e:
+            print(f"Error testing CrewAI: {e}")
+    
+    # Final Comparison Table
+    print("\n" + "="*70)
+    print("RESULTS SUMMARY")
+    print("="*70)
+    
+    if 'praisonai' in results:
+        baseline = results['praisonai']
+        
+        print(f"\n{'Framework':<25} {'Avg Time (μs)':<15} {'Avg Memory (KiB)':<18} {'Relative':<15}")
+        print("-" * 73)
+        
+        # Sort by time
+        sorted_results = sorted(results.items(), key=lambda x: x[1].avg_run_time)
+        
+        for name, result in sorted_results:
+            time_ratio = result.avg_run_time / baseline.avg_run_time if baseline.avg_run_time > 0 else 0
+            
+            time_str = f"{time_ratio:.2f}x"
+            
+            display_name = {
+                'agno': 'Agno',
+                'praisonai': 'PraisonAI',
+                'praisonai_litellm': 'PraisonAI (LiteLLM)',
+                'pydantic': 'PydanticAI',
+                'openai_agents': 'OpenAI Agents SDK',
+                'langgraph': 'LangGraph',
+                'crewai': 'CrewAI'
+            }.get(name, name)
+            
+            print(f"{display_name:<25} {result.avg_run_time*1e6:<15.2f} {result.avg_memory_usage*1024:<18.2f} {time_str:<15}")
+        
+        print("\n" + "="*70)

src/praisonai-agents/benchmarks/response_benchmark.py

@@ -0,0 +1,235 @@
+"""
+PraisonAI Agents - Response Time Benchmark
+
+Measures actual LLM response time with gpt-4o-mini across frameworks.
+
+Usage:
+    python benchmarks/response_benchmark.py
+"""
+
+import time
+import statistics
+import os
+from typing import List, Literal
+from dataclasses import dataclass, field
+
+# Ensure we use gpt-4o-mini
+os.environ['OPENAI_MODEL_NAME'] = 'gpt-4o-mini'
+
+
+@dataclass
+class ResponseResult:
+    """Response time measurement results"""
+    response_times: List[float] = field(default_factory=list)
+    
+    @property
+    def avg_time(self) -> float:
+        return statistics.mean(self.response_times) if self.response_times else 0
+    
+    @property
+    def min_time(self) -> float:
+        return min(self.response_times) if self.response_times else 0
+    
+    @property
+    def max_time(self) -> float:
+        return max(self.response_times) if self.response_times else 0
+    
+    @property
+    def std_dev(self) -> float:
+        return statistics.stdev(self.response_times) if len(self.response_times) > 1 else 0
+
+
+def get_weather(city: Literal["nyc", "sf"]):
+    """Use this to get weather information."""
+    if city == "nyc":
+        return "It might be cloudy in nyc"
+    elif city == "sf":
+        return "It's always sunny in sf"
+
+
+def benchmark_praisonai_response(iterations: int = 5) -> ResponseResult:
+    """Benchmark PraisonAI agent response time"""
+    from praisonaiagents import Agent
+    
+    result = ResponseResult()
+    
+    for i in range(iterations):
+        print(f"  PraisonAI iteration {i+1}/{iterations}...")
+        
+        # Create fresh agent each time
+        agent = Agent(
+            name="Test Agent",
+            instructions="Be concise, reply with one sentence.",
+            llm="gpt-4o-mini",
+            verbose=False
+        )
+        
+        start = time.perf_counter()
+        response = agent.start("What is the capital of France?")
+        end = time.perf_counter()
+        
+        result.response_times.append(end - start)
+        print(f"    Response time: {end - start:.3f}s")
+    
+    return result
+
+
+def benchmark_agno_response(iterations: int = 5) -> ResponseResult:
+    """Benchmark Agno agent response time"""
+    from agno.agent import Agent
+    from agno.models.openai import OpenAIChat
+    
+    result = ResponseResult()
+    
+    for i in range(iterations):
+        print(f"  Agno iteration {i+1}/{iterations}...")
+        
+        # Create fresh agent each time
+        agent = Agent(
+            model=OpenAIChat(id="gpt-4o-mini"),
+            system_message="Be concise, reply with one sentence.",
+        )
+        
+        start = time.perf_counter()
+        response = agent.run("What is the capital of France?")
+        end = time.perf_counter()
+        
+        result.response_times.append(end - start)
+        print(f"    Response time: {end - start:.3f}s")
+    
+    return result
+
+
+def benchmark_praisonai_with_tool(iterations: int = 5) -> ResponseResult:
+    """Benchmark PraisonAI agent response time with tool calling"""
+    from praisonaiagents import Agent
+    
+    result = ResponseResult()
+    
+    for i in range(iterations):
+        print(f"  PraisonAI (with tool) iteration {i+1}/{iterations}...")
+        
+        agent = Agent(
+            name="Weather Agent",
+            instructions="You are a weather assistant. Use the get_weather tool to answer questions.",
+            llm="gpt-4o-mini",
+            tools=[get_weather],
+            verbose=False
+        )
+        
+        start = time.perf_counter()
+        response = agent.start("What's the weather in NYC?")
+        end = time.perf_counter()
+        
+        result.response_times.append(end - start)
+        print(f"    Response time: {end - start:.3f}s")
+    
+    return result
+
+
+def benchmark_agno_with_tool(iterations: int = 5) -> ResponseResult:
+    """Benchmark Agno agent response time with tool calling"""
+    from agno.agent import Agent
+    from agno.models.openai import OpenAIChat
+    
+    result = ResponseResult()
+    
+    for i in range(iterations):
+        print(f"  Agno (with tool) iteration {i+1}/{iterations}...")
+        
+        agent = Agent(
+            model=OpenAIChat(id="gpt-4o-mini"),
+            system_message="You are a weather assistant. Use the get_weather tool to answer questions.",
+            tools=[get_weather],
+        )
+        
+        start = time.perf_counter()
+        response = agent.run("What's the weather in NYC?")
+        end = time.perf_counter()
+        
+        result.response_times.append(end - start)
+        print(f"    Response time: {end - start:.3f}s")
+    
+    return result
+
+
+if __name__ == "__main__":
+    print("="*70)
+    print("PraisonAI Agents - Response Time Benchmark")
+    print("="*70)
+    print("\nModel: gpt-4o-mini")
+    print("Measures end-to-end response time including:")
+    print("  - Agent instantiation")
+    print("  - API call to OpenAI")
+    print("  - Response processing")
+    
+    iterations = 5
+    results = {}
+    
+    # Test 1: Simple response (no tools)
+    print("\n" + "="*70)
+    print("TEST 1: Simple Response (No Tools)")
+    print("="*70)
+    
+    print("\nBenchmarking PraisonAI...")
+    try:
+        results['praisonai_simple'] = benchmark_praisonai_response(iterations)
+    except Exception as e:
+        print(f"Error: {e}")
+        import traceback
+        traceback.print_exc()
+    
+    print("\nBenchmarking Agno...")
+    try:
+        results['agno_simple'] = benchmark_agno_response(iterations)
+    except Exception as e:
+        print(f"Error: {e}")
+        import traceback
+        traceback.print_exc()
+    
+    # Test 2: Response with tool calling
+    print("\n" + "="*70)
+    print("TEST 2: Response with Tool Calling")
+    print("="*70)
+    
+    print("\nBenchmarking PraisonAI with tool...")
+    try:
+        results['praisonai_tool'] = benchmark_praisonai_with_tool(iterations)
+    except Exception as e:
+        print(f"Error: {e}")
+        import traceback
+        traceback.print_exc()
+    
+    print("\nBenchmarking Agno with tool...")
+    try:
+        results['agno_tool'] = benchmark_agno_with_tool(iterations)
+    except Exception as e:
+        print(f"Error: {e}")
+        import traceback
+        traceback.print_exc()
+    
+    # Summary
+    print("\n" + "="*70)
+    print("SUMMARY RESULTS")
+    print("="*70)
+    
+    print(f"\n{'Test':<30} {'PraisonAI (avg)':<20} {'Agno (avg)':<20} {'Ratio':<15}")
+    print("-" * 85)
+    
+    if 'praisonai_simple' in results and 'agno_simple' in results:
+        p_time = results['praisonai_simple'].avg_time
+        a_time = results['agno_simple'].avg_time
+        ratio = p_time / a_time if a_time > 0 else float('inf')
+        print(f"{'Simple Response':<30} {p_time:<20.3f}s {a_time:<20.3f}s {ratio:.2f}x")
+    
+    if 'praisonai_tool' in results and 'agno_tool' in results:
+        p_time = results['praisonai_tool'].avg_time
+        a_time = results['agno_tool'].avg_time
+        ratio = p_time / a_time if a_time > 0 else float('inf')
+        print(f"{'With Tool Calling':<30} {p_time:<20.3f}s {a_time:<20.3f}s {ratio:.2f}x")
+    
+    print("\n" + "="*70)
+    print("NOTE")
+    print("="*70)
+    print("\nResponse time is dominated by OpenAI API latency.")
+    print("Agent instantiation overhead is negligible in real-world usage.")

src/praisonai-agents/benchmarks/simple_benchmark.py

@@ -0,0 +1,103 @@
+#!/usr/bin/env python3
+"""
+PraisonAI Agents - Performance Benchmark
+
+Compares agent instantiation times across popular AI agent frameworks.
+
+Usage:
+    python benchmarks/simple_benchmark.py
+"""
+
+import time
+from typing import Literal
+
+
+ITERATIONS = 100
+
+
+def sample_tool(city: Literal['nyc', 'sf']):
+    """Sample tool for benchmark testing."""
+    if city == 'nyc':
+        return 'cloudy'
+    return 'sunny'
+
+
+def measure_instantiation(create_fn, iterations=ITERATIONS):
+    """Measure average instantiation time in microseconds."""
+    times = []
+    for _ in range(iterations):
+        start = time.perf_counter()
+        create_fn()
+        times.append((time.perf_counter() - start) * 1_000_000)
+    return sum(times) / len(times)
+
+
+def run_benchmark():
+    """Run the benchmark across all available frameworks."""
+    tools = [sample_tool]
+    results = {}
+    
+    print('=' * 60)
+    print('PraisonAI Agents - Performance Benchmark')
+    print('=' * 60)
+    print(f'\nIterations: {ITERATIONS}')
+    print('Metric: Agent instantiation time (microseconds)\n')
+    
+    # PraisonAI
+    print("Testing PraisonAI...")
+    from praisonaiagents import Agent as PraisonAgent
+    results['PraisonAI'] = measure_instantiation(
+        lambda: PraisonAgent(name='Test', llm='gpt-4o-mini', tools=tools, verbose=False)
+    )
+    
+    results['PraisonAI (LiteLLM)'] = measure_instantiation(
+        lambda: PraisonAgent(name='Test', llm='openai/gpt-4o-mini', tools=tools, verbose=False)
+    )
+    
+    # Other frameworks for comparison
+    print("Testing other frameworks...")
+    
+    try:
+        from agno.agent import Agent as AgnoAgent
+        from agno.models.openai import OpenAIChat
+        results['Agno'] = measure_instantiation(
+            lambda: AgnoAgent(model=OpenAIChat(id='gpt-4o-mini'), tools=tools)
+        )
+    except ImportError:
+        pass
+    
+    try:
+        from pydantic_ai import Agent as PydanticAgent
+        results['PydanticAI'] = measure_instantiation(
+            lambda: PydanticAgent('openai:gpt-4o-mini')
+        )
+    except ImportError:
+        pass
+    
+    try:
+        from agents import Agent as OpenAIAgent
+        results['OpenAI Agents SDK'] = measure_instantiation(
+            lambda: OpenAIAgent(name='Test', model='gpt-4o-mini')
+        )
+    except ImportError:
+        pass
+    
+    # Print results
+    print('\n' + '=' * 60)
+    print('RESULTS')
+    print('=' * 60)
+    
+    baseline = results.get('PraisonAI', 1)
+    print(f"\n{'Framework':<25} {'Avg Time (μs)':<15} {'Relative':<10}")
+    print('-' * 50)
+    
+    for name, avg in sorted(results.items(), key=lambda x: x[1]):
+        ratio = avg / baseline
+        print(f'{name:<25} {avg:<15.2f} {ratio:.2f}x')
+    
+    print('\n' + '=' * 60)
+    return results
+
+
+if __name__ == '__main__':
+    run_benchmark()

src/praisonai-agents/callback_test_log.txt

@@ -0,0 +1,9 @@
+Starting test with verbose=False
+CALLBACK EXECUTED!
+Message: 
+You need to do the following task: Say hello world.
+Expected Output: A greeting.
+Please provide only the final result of your work. Do not add any conversation or extra explanation.
+Response: Hello, World!
+Other kwargs: {'markdown': True, 'generation_time': 2.685833215713501, 'agent_name': 'TestAgent', 'agent_role': 'Assistant', 'agent_tools': None, 'task_name': 'simple_task', 'task_description': 'Say hello world', 'task_id': 0}
+--------------------------------------------------

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/examples/plugin_template/README.md

@@ -0,0 +1,94 @@
+# PraisonAI Example Plugin
+
+This is a template for creating tool plugins for PraisonAI Agents.
+
+## Installation
+
+```bash
+# Install in development mode
+pip install -e .
+
+# Or install from PyPI (after publishing)
+pip install praisonai-example-plugin
+```
+
+## Usage
+
+After installation, tools are automatically discovered:
+
+```python
+from praisonaiagents import Agent
+
+# Use tools by name - they're auto-discovered!
+agent = Agent(
+    name="Assistant",
+    tools=["example_greet", "example_math", "example_reverse"]
+)
+
+# Or import and use directly
+from praisonai_example_plugin import GreetTool, MathTool, reverse_text
+
+agent = Agent(
+    name="Assistant", 
+    tools=[GreetTool(), MathTool(), reverse_text]
+)
+```
+
+## Creating Your Own Plugin
+
+1. **Copy this template**
+2. **Rename the package** in `pyproject.toml` and `src/` directory
+3. **Add your tools** following the patterns in `__init__.py`
+4. **Update entry_points** in `pyproject.toml` to register your tools
+
+### Tool Patterns
+
+**Class-based tool:**
+```python
+from praisonaiagents import BaseTool
+
+class MyTool(BaseTool):
+    name = "my_tool"
+    description = "What this tool does"
+    
+    def run(self, param: str) -> str:
+        return f"Result: {param}"
+```
+
+**Decorator-based tool:**
+```python
+from praisonaiagents import tool
+
+@tool
+def my_function(param: str) -> str:
+    """What this function does."""
+    return f"Result: {param}"
+```
+
+### Registering Tools
+
+In `pyproject.toml`:
+```toml
+[project.entry-points."praisonaiagents.tools"]
+my_tool = "my_package:MyTool"
+my_function = "my_package:my_function"
+```
+
+## Testing
+
+```bash
+pip install -e ".[dev]"
+pytest
+```
+
+## Publishing
+
+```bash
+pip install build twine
+python -m build
+twine upload dist/*
+```
+
+## License
+
+MIT

src/praisonai-agents/examples/plugin_template/pyproject.toml

@@ -0,0 +1,49 @@
+[build-system]
+requires = ["setuptools>=61.0", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "praisonai-example-plugin"
+version = "1.0.0"
+description = "Example plugin for PraisonAI Agents - demonstrates tool plugin development"
+readme = "README.md"
+license = {text = "MIT"}
+requires-python = ">=3.10"
+authors = [
+    {name = "Your Name", email = "[email protected]"}
+]
+keywords = ["praisonai", "agents", "tools", "plugin"]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+]
+
+dependencies = [
+    "praisonaiagents>=0.0.168",
+]
+
+[project.optional-dependencies]
+dev = [
+    "pytest>=7.0.0",
+    "pytest-mock>=3.10.0",
+]
+
+# This is the key part - register your tools as entry points
+# The format is: tool_name = "module.path:ClassName" or "module.path:function_name"
+[project.entry-points."praisonaiagents.tools"]
+example_greet = "praisonai_example_plugin:GreetTool"
+example_math = "praisonai_example_plugin:MathTool"
+example_reverse = "praisonai_example_plugin:reverse_text"
+
+[project.urls]
+Homepage = "https://github.com/yourusername/praisonai-example-plugin"
+Documentation = "https://github.com/yourusername/praisonai-example-plugin#readme"
+Repository = "https://github.com/yourusername/praisonai-example-plugin"
+
+[tool.setuptools.packages.find]
+where = ["src"]

src/praisonai-agents/examples/plugin_template/src/praisonai_example_plugin/__init__.py

@@ -0,0 +1,119 @@
+"""PraisonAI Example Plugin - Demonstrates how to create tool plugins.
+
+This package shows how external developers can create pip-installable
+tool plugins for PraisonAI Agents.
+
+Usage after installation:
+    from praisonaiagents import Agent
+    
+    # Tools are auto-discovered via entry_points!
+    agent = Agent(tools=["example_greet", "example_math"])
+    
+    # Or import directly
+    from praisonai_example_plugin import GreetTool, MathTool
+    agent = Agent(tools=[GreetTool(), MathTool()])
+"""
+
+from typing import Dict, Any
+from praisonaiagents import BaseTool, tool
+
+
+# =============================================================================
+# Class-Based Tools
+# =============================================================================
+
+class GreetTool(BaseTool):
+    """A tool that generates personalized greetings.
+    
+    This demonstrates a simple class-based tool with multiple parameters.
+    """
+    name = "example_greet"
+    description = "Generate a personalized greeting message"
+    version = "1.0.0"
+    
+    def run(self, name: str, greeting: str = "Hello", punctuation: str = "!") -> str:
+        """Generate a greeting.
+        
+        Args:
+            name: The name of the person to greet
+            greeting: The greeting word (default: "Hello")
+            punctuation: End punctuation (default: "!")
+            
+        Returns:
+            A formatted greeting string
+        """
+        return f"{greeting}, {name}{punctuation}"
+
+
+class MathTool(BaseTool):
+    """A tool for basic mathematical operations.
+    
+    Demonstrates returning structured data from a tool.
+    """
+    name = "example_math"
+    description = "Perform basic math operations: add, subtract, multiply, divide"
+    version = "1.0.0"
+    
+    def run(
+        self,
+        a: float,
+        b: float,
+        operation: str = "add"
+    ) -> Dict[str, Any]:
+        """Perform a math operation.
+        
+        Args:
+            a: First number
+            b: Second number  
+            operation: One of 'add', 'subtract', 'multiply', 'divide'
+            
+        Returns:
+            Dict containing result and operation details
+        """
+        ops = {
+            "add": ("+", lambda x, y: x + y),
+            "subtract": ("-", lambda x, y: x - y),
+            "multiply": ("*", lambda x, y: x * y),
+            "divide": ("/", lambda x, y: x / y if y != 0 else float('inf')),
+        }
+        
+        if operation not in ops:
+            return {"error": f"Unknown operation: {operation}. Use: {list(ops.keys())}"}
+        
+        symbol, func = ops[operation]
+        result = func(a, b)
+        
+        return {
+            "result": result,
+            "expression": f"{a} {symbol} {b} = {result}",
+            "operation": operation
+        }
+
+
+# =============================================================================
+# Decorator-Based Tools
+# =============================================================================
+
+@tool(name="example_reverse", description="Reverse a string of text")
+def reverse_text(text: str) -> str:
+    """Reverse the input text.
+    
+    Args:
+        text: The text to reverse
+        
+    Returns:
+        The reversed text
+    """
+    return text[::-1]
+
+
+# =============================================================================
+# Package Exports
+# =============================================================================
+
+__version__ = "1.0.0"
+__all__ = [
+    "GreetTool",
+    "MathTool", 
+    "reverse_text",
+]

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

@@ -26,9 +26,22 @@ from .agent.deep_research_agent import (
     FileSearchCall,
     Provider
 )
+from .agent.query_rewriter_agent import (
+    QueryRewriterAgent,
+    RewriteStrategy,
+    RewriteResult
+)
+from .agent.prompt_expander_agent import (
+    PromptExpanderAgent,
+    ExpandStrategy,
+    ExpandResult
+)
 from .agents.agents import PraisonAIAgents
 from .task.task import Task
 from .tools.tools import Tools
+from .tools.base import BaseTool, ToolResult, ToolValidationError, validate_tool
+from .tools.decorator import tool, FunctionTool
+from .tools.registry import get_registry, register_tool, get_tool, ToolRegistry
 from .agents.autoagents import AutoAgents
 from .knowledge.knowledge import Knowledge
 from .knowledge.chunking import Chunking
@@ -41,7 +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:
@@ -165,6 +228,17 @@ __all__ = [
     'Agents',
     'Tools',
     'Task',
+    # Plugin system
+    'BaseTool',
+    'ToolResult',
+    'ToolValidationError',
+    'validate_tool',
+    'tool',
+    'FunctionTool',
+    'ToolRegistry',
+    'get_registry',
+    'register_tool',
+    'get_tool',
     'TaskOutput',
     'ReflectionOutput',
     'AutoAgents',
@@ -206,7 +280,13 @@ __all__ = [
     'CodeExecutionStep',
     'MCPCall',
     'FileSearchCall',
-    'Provider'
+    'Provider',
+    'QueryRewriterAgent',
+    'RewriteStrategy',
+    'RewriteResult',
+    'PromptExpanderAgent',
+    'ExpandStrategy',
+    'ExpandResult'
 ]
 
 # Add MCP to __all__ if available
@@ -217,3 +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/__init__.py

@@ -15,6 +15,16 @@ from .deep_research_agent import (
     FileSearchCall,
     Provider
 )
+from .query_rewriter_agent import (
+    QueryRewriterAgent,
+    RewriteStrategy,
+    RewriteResult
+)
+from .prompt_expander_agent import (
+    PromptExpanderAgent,
+    ExpandStrategy,
+    ExpandResult
+)
 
 __all__ = [
     'Agent',
@@ -35,5 +45,11 @@ __all__ = [
     'CodeExecutionStep',
     'MCPCall',
     'FileSearchCall',
-    'Provider'
+    'Provider',
+    'QueryRewriterAgent',
+    'RewriteStrategy',
+    'RewriteResult',
+    'PromptExpanderAgent',
+    'ExpandStrategy',
+    'ExpandResult'
 ]

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
                 
@@ -455,6 +479,15 @@ class Agent:
         if callable(tools):
             # If a single function/callable is passed, wrap it in a list
             self.tools = [tools]
+        elif isinstance(tools, str):
+            # Single tool name string - resolve from registry
+            self.tools = self._resolve_tool_names([tools])
+        elif isinstance(tools, (list, tuple)):
+            # Check if list contains strings (tool names) that need resolution
+            if tools and all(isinstance(t, str) for t in tools):
+                self.tools = self._resolve_tool_names(tools)
+            else:
+                self.tools = list(tools)
         else:
             # Handle all falsy values (None, False, 0, "", etc.) by defaulting to empty list
             self.tools = tools or []
@@ -462,7 +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
@@ -502,6 +536,37 @@ 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)
+        # NOTE: Lazy initialization - rules are loaded only when accessed (performance optimization)
+        self._rules_manager = None
+        self._rules_manager_initialized = False
+        
+        # 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
@@ -533,6 +598,15 @@ 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  # Store raw, resolve lazily
+        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."""
@@ -541,6 +615,17 @@ Your Goal: {self.goal}
             self._console = Console()
         return self._console
     
+    @property
+    def fast_context_path(self):
+        """Lazily resolve fast_context_path - avoids os.getcwd() on every init."""
+        if self._fast_context_path is None:
+            self._fast_context_path = os.getcwd()
+        return self._fast_context_path
+    
+    @fast_context_path.setter
+    def fast_context_path(self, value):
+        self._fast_context_path = value
+    
     @property
     def _openai_client(self):
         """Lazily initialize OpenAI client only when needed."""
@@ -567,6 +652,363 @@ 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
+    
+    @property
+    def rules_manager(self):
+        """
+        Lazy-initialized RulesManager for persistent rules/instructions.
+        
+        This property initializes the RulesManager only when first accessed,
+        avoiding expensive filesystem operations during agent instantiation.
+        
+        Returns:
+            RulesManager instance or None if not available
+        """
+        if not self._rules_manager_initialized:
+            self._init_rules_manager()
+        return self._rules_manager
+    
+    def _init_rules_manager(self):
+        """
+        Initialize RulesManager for persistent rules/instructions.
+        
+        Automatically discovers rules from:
+        - ~/.praison/rules/ (global)
+        - .praison/rules/ (workspace)
+        - Subdirectory rules
+        
+        NOTE: This is called lazily via the rules_manager property for performance.
+        """
+        self._rules_manager_initialized = True
+        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.
@@ -813,17 +1255,35 @@ 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 (lazy initialization)
+        if self._rules_manager_initialized and 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
@@ -853,9 +1313,9 @@ 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
 
@@ -1008,6 +1468,31 @@ Your Goal: {self.goal}"""
             tools=self.tools
         )
 
+    def _resolve_tool_names(self, tool_names):
+        """Resolve tool names to actual tool instances from registry.
+        
+        Args:
+            tool_names: List of tool name strings
+            
+        Returns:
+            List of resolved tool instances
+        """
+        resolved = []
+        try:
+            from ..tools.registry import get_registry
+            registry = get_registry()
+            
+            for name in tool_names:
+                tool = registry.get(name)
+                if tool is not None:
+                    resolved.append(tool)
+                else:
+                    logging.warning(f"Tool '{name}' not found in registry")
+        except ImportError:
+            logging.warning("Tool registry not available, cannot resolve tool names")
+        
+        return resolved
+
     def _cast_arguments(self, func, arguments):
         """Cast arguments to their expected types based on function signature."""
         if not callable(func) or not arguments:
@@ -1084,34 +1569,83 @@ Your Goal: {self.goal}"""
         # Special handling for MCP tools
         # Check if tools is an MCP instance with the requested function name
         from ..mcp.mcp import MCP
+        
+        # Helper function to execute MCP tool
+        def _execute_mcp_tool(mcp_instance, func_name, args):
+            """Execute a tool from an MCP instance."""
+            # Handle SSE MCP client
+            if hasattr(mcp_instance, 'is_sse') and mcp_instance.is_sse:
+                if hasattr(mcp_instance, 'sse_client'):
+                    for tool in mcp_instance.sse_client.tools:
+                        if tool.name == func_name:
+                            logging.debug(f"Found matching SSE MCP tool: {func_name}")
+                            return True, tool(**args)
+            # Handle HTTP Stream MCP client
+            if hasattr(mcp_instance, 'is_http_stream') and mcp_instance.is_http_stream:
+                if hasattr(mcp_instance, 'http_stream_client'):
+                    for tool in mcp_instance.http_stream_client.tools:
+                        if tool.name == func_name:
+                            logging.debug(f"Found matching HTTP Stream MCP tool: {func_name}")
+                            return True, tool(**args)
+            # Handle WebSocket MCP client
+            if hasattr(mcp_instance, 'is_websocket') and mcp_instance.is_websocket:
+                if hasattr(mcp_instance, 'websocket_client'):
+                    for tool in mcp_instance.websocket_client.tools:
+                        if tool.name == func_name:
+                            logging.debug(f"Found matching WebSocket MCP tool: {func_name}")
+                            return True, tool(**args)
+            # Handle stdio MCP client
+            if hasattr(mcp_instance, 'runner'):
+                for mcp_tool in mcp_instance.runner.tools:
+                    if hasattr(mcp_tool, 'name') and mcp_tool.name == func_name:
+                        logging.debug(f"Found matching MCP tool: {func_name}")
+                        return True, mcp_instance.runner.call_tool(func_name, args)
+            return False, None
+        
+        # Check if tools is a single MCP instance
         if isinstance(self.tools, MCP):
             logging.debug(f"Looking for MCP tool {function_name}")
-            
-            # Handle SSE MCP client
-            if hasattr(self.tools, 'is_sse') and self.tools.is_sse:
-                if hasattr(self.tools, 'sse_client'):
-                    for tool in self.tools.sse_client.tools:
-                        if tool.name == function_name:
-                            logging.debug(f"Found matching SSE MCP tool: {function_name}")
-                            return tool(**arguments)
-            # Handle stdio MCP client
-            elif hasattr(self.tools, 'runner'):
-                # Check if any of the MCP tools match the function name
-                for mcp_tool in self.tools.runner.tools:
-                    if hasattr(mcp_tool, 'name') and mcp_tool.name == function_name:
-                        logging.debug(f"Found matching MCP tool: {function_name}")
-                        return self.tools.runner.call_tool(function_name, arguments)
+            found, result = _execute_mcp_tool(self.tools, function_name, arguments)
+            if found:
+                return result
+        
+        # Check if tools is a list that may contain MCP instances
+        if isinstance(self.tools, (list, tuple)):
+            for tool in self.tools:
+                if isinstance(tool, MCP):
+                    logging.debug(f"Looking for MCP tool {function_name} in MCP instance")
+                    found, result = _execute_mcp_tool(tool, function_name, arguments)
+                    if found:
+                        return result
 
         # Try to find the function in the agent's tools list first
         func = None
         for tool in self.tools if isinstance(self.tools, (list, tuple)) else []:
+            # Check for BaseTool instances (plugin system)
+            from ..tools.base import BaseTool
+            if isinstance(tool, BaseTool) and tool.name == function_name:
+                func = tool
+                break
+            # Check for FunctionTool (decorated functions)
+            if hasattr(tool, 'name') and getattr(tool, 'name', None) == function_name:
+                func = tool
+                break
             if (callable(tool) and getattr(tool, '__name__', '') == function_name) or \
                (inspect.isclass(tool) and tool.__name__ == function_name):
                 func = tool
                 break
         
         if func is None:
-            # If not found in tools, try globals and main
+            # Check the global tool registry for plugins
+            try:
+                from ..tools.registry import get_registry
+                registry = get_registry()
+                func = registry.get(function_name)
+            except ImportError:
+                pass
+        
+        if func is None:
+            # If not found in tools or registry, try globals and main
             func = globals().get(function_name)
             if not func:
                 import __main__
@@ -1119,6 +1653,12 @@ Your Goal: {self.goal}"""
 
         if func:
             try:
+                # BaseTool instances (plugin system) - call run() method
+                from ..tools.base import BaseTool
+                if isinstance(func, BaseTool):
+                    casted_arguments = self._cast_arguments(func.run, arguments)
+                    return func.run(**casted_arguments)
+                
                 # Langchain: If it's a class with run but not _run, instantiate and call run
                 if inspect.isclass(func) and hasattr(func, 'run') and not hasattr(func, '_run'):
                     instance = func()
@@ -1391,7 +1931,8 @@ Your Goal: {self.goal}"""
                 if tool_param is not None:
                     from ..mcp.mcp import MCP
                     if isinstance(tool_param, MCP) and hasattr(tool_param, 'to_openai_tool'):
-                        logging.debug("Converting MCP tool to OpenAI format")
+                        # Single MCP instance
+                        logging.debug("Converting single MCP tool to OpenAI format")
                         openai_tool = tool_param.to_openai_tool()
                         if openai_tool:
                             # Handle both single tool and list of tools
@@ -1400,6 +1941,22 @@ Your Goal: {self.goal}"""
                             else:
                                 tool_param = [openai_tool]
                             logging.debug(f"Converted MCP tool: {tool_param}")
+                    elif isinstance(tool_param, (list, tuple)):
+                        # List that may contain MCP instances - convert each MCP to OpenAI format
+                        converted_tools = []
+                        for t in tool_param:
+                            if isinstance(t, MCP) and hasattr(t, 'to_openai_tool'):
+                                logging.debug("Converting MCP instance in list to OpenAI format")
+                                openai_tools = t.to_openai_tool()
+                                if isinstance(openai_tools, list):
+                                    converted_tools.extend(openai_tools)
+                                elif openai_tools:
+                                    converted_tools.append(openai_tools)
+                            else:
+                                # Keep non-MCP tools as-is
+                                converted_tools.append(t)
+                        tool_param = converted_tools
+                        logging.debug(f"Converted {len(converted_tools)} tools from list")
                 
                 # Store chat history length for potential rollback
                 chat_history_length = len(self.chat_history)
@@ -2092,21 +2649,189 @@ 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/deep_research_agent.py

@@ -254,7 +254,17 @@ class DeepResearchAgent:
             max_wait_time: Maximum seconds to wait for research completion
         """
         self.name = name or "DeepResearchAgent"
-        self.instructions = instructions
+        # Default instructions for comprehensive research
+        default_instructions = """
+You are a professional research analyst. When conducting research:
+- Include specific figures, trends, statistics, and measurable outcomes
+- Prioritize reliable, up-to-date sources: peer-reviewed research, official organizations, regulatory agencies
+- Include inline citations and return all source metadata
+- Be analytical, avoid generalities, and ensure data-backed reasoning
+- Structure your response with clear headers and formatting
+- If creating comparisons, use tables for clarity
+"""
+        self.instructions = instructions or default_instructions
         self.verbose = verbose
         self.summary_mode = summary_mode
         self.enable_web_search = enable_web_search

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

@@ -0,0 +1,415 @@
+"""
+Prompt Expander Agent Module
+
+This module provides the PromptExpanderAgent class for expanding short prompts
+into detailed, comprehensive prompts for better task execution.
+
+Unlike QueryRewriterAgent (which optimizes queries for search/retrieval),
+PromptExpanderAgent focuses on enriching prompts for task execution.
+
+Supported Expansion Strategies:
+- **BASIC**: Simple expansion with clarity improvements
+- **DETAILED**: Rich expansion with context, constraints, and examples
+- **STRUCTURED**: Expansion with clear structure (task, format, requirements)
+- **CREATIVE**: Expansion with creative flair and vivid language
+- **AUTO**: Automatically selects the best strategy based on prompt analysis
+
+Example:
+    from praisonaiagents import PromptExpanderAgent, ExpandStrategy
+    
+    agent = PromptExpanderAgent(verbose=True)
+    
+    # Basic expansion
+    result = agent.expand("write a movie script in 3 lines")
+    print(result.expanded_prompt)
+    
+    # Detailed expansion
+    result = agent.expand("blog about AI", strategy=ExpandStrategy.DETAILED)
+    print(result.expanded_prompt)
+    
+    # With tools for context gathering
+    from praisonaiagents.tools import internet_search
+    agent = PromptExpanderAgent(tools=[internet_search], verbose=True)
+    result = agent.expand("latest AI developments")
+"""
+
+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."""
+    BASIC = "basic"
+    DETAILED = "detailed"
+    STRUCTURED = "structured"
+    CREATIVE = "creative"
+    AUTO = "auto"
+
+
+@dataclass
+class ExpandResult:
+    """Result of a prompt expansion operation."""
+    original_prompt: str
+    expanded_prompt: str
+    strategy_used: ExpandStrategy
+    metadata: Dict[str, Any] = field(default_factory=dict)
+    
+    def __repr__(self):
+        return f"ExpandResult(strategy={self.strategy_used.value}, original_len={len(self.original_prompt)}, expanded_len={len(self.expanded_prompt)})"
+
+
+class PromptExpanderAgent:
+    """
+    Agent for expanding short prompts into detailed, comprehensive prompts.
+    
+    This agent transforms brief task descriptions into rich, detailed prompts
+    that provide better context and guidance for task execution.
+    
+    Key difference from QueryRewriterAgent:
+    - QueryRewriterAgent: Optimizes queries for search/retrieval (RAG)
+    - PromptExpanderAgent: Expands prompts for detailed task execution
+    
+    Attributes:
+        name: Name of the agent
+        model: LLM model to use for expansion
+        verbose: Whether to print detailed logs
+        temperature: Temperature for LLM generation (higher = more creative)
+        tools: Optional tools for context gathering
+        
+    Example:
+        agent = PromptExpanderAgent(model="gpt-4o-mini", verbose=True)
+        result = agent.expand("write a poem", strategy=ExpandStrategy.CREATIVE)
+        print(result.expanded_prompt)
+    """
+    
+    # Default prompts for each strategy
+    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:
+- Keep it as a TASK/COMMAND (do NOT convert to a question)
+- 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}
+
+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:
+- Keep it as a TASK/COMMAND (do NOT convert to a question)
+- Add specific details about what to include
+- Specify quality expectations
+- 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}
+
+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:
+- Keep it as a TASK/COMMAND (do NOT convert to a question)
+- Use a clear structure:
+  * Task: What to do
+  * Format: Expected output format
+  * Requirements: Specific requirements
+  * Style: Tone or style guidance
+  * Constraints: Any limitations
+- Be specific and actionable
+- Use the current date for any time-sensitive context
+
+Original prompt: {prompt}
+
+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:
+- Keep it as a TASK/COMMAND (do NOT convert to a question)
+- Use evocative, inspiring language
+- Paint a picture of the desired outcome
+- 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}
+
+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
+For task prompts: Preserve the action-oriented nature
+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}
+
+Expanded prompt:"""
+    }
+    
+    def __init__(
+        self,
+        name: str = "PromptExpanderAgent",
+        model: str = "gpt-4o-mini",
+        instructions: Optional[str] = None,
+        verbose: bool = False,
+        temperature: float = 0.7,  # Higher for creativity
+        max_tokens: int = 1000,
+        tools: Optional[List[Any]] = None
+    ):
+        """
+        Initialize the PromptExpanderAgent.
+        
+        Args:
+            name: Name of the agent
+            model: LLM model to use (default: gpt-4o-mini)
+            instructions: Custom instructions for the agent
+            verbose: Whether to print detailed logs
+            temperature: Temperature for LLM generation (default: 0.7 for creativity)
+            max_tokens: Maximum tokens for LLM response
+            tools: Optional list of tools for context gathering
+        """
+        self.name = name
+        self.model = model
+        self.instructions = instructions
+        self.verbose = verbose
+        self.temperature = temperature
+        self.max_tokens = max_tokens
+        self.tools = tools or []
+        self._agent = None  # Lazy initialized
+        
+        # Setup logging
+        self.logger = logging.getLogger(__name__)
+        if self.verbose:
+            self.logger.setLevel(logging.DEBUG)
+    
+    @property
+    def agent(self):
+        """Lazy initialization of internal Agent."""
+        if self._agent is None:
+            from .agent import Agent
+            self._agent = Agent(
+                name=self.name,
+                role="Prompt Expansion Specialist",
+                goal="Expand brief prompts into detailed, actionable prompts for better task execution. Use available tools when needed to gather context.",
+                backstory="You are an expert at understanding user intent and transforming brief requests into comprehensive, detailed prompts.",
+                tools=self.tools,
+                llm=self.model,
+                verbose=self.verbose,
+                markdown=False
+            )
+        return self._agent
+    
+    def _call_agent(self, prompt: str) -> str:
+        """Call the internal Agent with the given prompt."""
+        try:
+            response = self.agent.chat(prompt)
+            return response if response else ""
+        except Exception as e:
+            self.logger.error(f"Agent call failed: {e}")
+            raise
+    
+    def _is_short_prompt(self, prompt: str, threshold: int = 10) -> bool:
+        """Check if prompt is considered short."""
+        return len(prompt.split()) < threshold
+    
+    def _is_task_prompt(self, prompt: str) -> bool:
+        """Check if prompt is a task/command (vs a question)."""
+        task_verbs = [
+            "write", "create", "generate", "build", "make", "design",
+            "develop", "implement", "compose", "draft", "produce",
+            "craft", "construct", "prepare", "draw", "paint", "code"
+        ]
+        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 (non-technical)."""
+        creative_indicators = [
+            "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)
+    
+    def _detect_strategy(self, prompt: str) -> ExpandStrategy:
+        """Auto-detect the best expansion strategy for a prompt."""
+        # Very short prompts need detailed expansion
+        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
+        
+        # Task prompts get structured expansion
+        if self._is_task_prompt(prompt):
+            return ExpandStrategy.STRUCTURED
+        
+        # Default to basic
+        return ExpandStrategy.BASIC
+    
+    def expand(
+        self,
+        prompt: str,
+        strategy: ExpandStrategy = ExpandStrategy.AUTO,
+        context: Optional[str] = None
+    ) -> ExpandResult:
+        """
+        Expand a prompt using the specified strategy.
+        
+        All operations go through an internal Agent which automatically handles
+        tool calling when tools are provided. The agent decides when to use tools
+        based on the prompt context.
+        
+        Args:
+            prompt: The original user prompt
+            strategy: Expansion strategy to use
+            context: Additional context to help with expansion
+            
+        Returns:
+            ExpandResult containing the expanded prompt
+        """
+        if self.verbose:
+            self.logger.debug(f"Expanding prompt: '{prompt}' with strategy: {strategy.value}")
+            if self.tools:
+                print(f"[cyan]Agent has {len(self.tools)} tools available (will use if needed)...[/cyan]")
+        
+        # Auto-detect strategy if needed
+        if strategy == ExpandStrategy.AUTO:
+            strategy = self._detect_strategy(prompt)
+            if self.verbose:
+                self.logger.debug(f"Auto-detected strategy: {strategy.value}")
+        
+        # Build the expansion prompt
+        if strategy == ExpandStrategy.AUTO:
+            # 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 "",
+            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()
+        
+        # If expansion failed, return original
+        if not expanded:
+            expanded = prompt
+        
+        return ExpandResult(
+            original_prompt=prompt,
+            expanded_prompt=expanded,
+            strategy_used=strategy,
+            metadata={
+                "original_length": len(prompt),
+                "expanded_length": len(expanded),
+                "has_context": bool(context),
+                "has_tools": bool(self.tools)
+            }
+        )
+    
+    # Convenience methods for direct strategy access
+    def expand_basic(self, prompt: str, context: Optional[str] = None) -> ExpandResult:
+        """Convenience method for basic expansion."""
+        return self.expand(prompt, strategy=ExpandStrategy.BASIC, context=context)
+    
+    def expand_detailed(self, prompt: str, context: Optional[str] = None) -> ExpandResult:
+        """Convenience method for detailed expansion."""
+        return self.expand(prompt, strategy=ExpandStrategy.DETAILED, context=context)
+    
+    def expand_structured(self, prompt: str, context: Optional[str] = None) -> ExpandResult:
+        """Convenience method for structured expansion."""
+        return self.expand(prompt, strategy=ExpandStrategy.STRUCTURED, context=context)
+    
+    def expand_creative(self, prompt: str, context: Optional[str] = None) -> ExpandResult:
+        """Convenience method for creative expansion."""
+        return self.expand(prompt, strategy=ExpandStrategy.CREATIVE, context=context)

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

@@ -0,0 +1,632 @@
+"""
+Query Rewriter Agent Module
+
+This module provides the QueryRewriterAgent class for transforming user queries
+to improve retrieval quality in RAG applications.
+
+Supported Rewriting Strategies:
+- **BASIC**: Simple rephrasing for clarity and keyword optimization
+- **HYDE**: Hypothetical Document Embeddings - generates a hypothetical answer
+- **STEP_BACK**: Generates higher-level concept questions for complex queries
+- **SUB_QUERIES**: Decomposes multi-part questions into focused sub-queries
+- **MULTI_QUERY**: Generates multiple paraphrased versions for ensemble retrieval
+- **CONTEXTUAL**: Uses conversation history to resolve references and context
+
+Example:
+    from praisonaiagents import QueryRewriterAgent, RewriteStrategy
+    
+    agent = QueryRewriterAgent(verbose=True)
+    
+    # Basic rewriting
+    result = agent.rewrite("AI trends")
+    print(result.rewritten_queries)
+    
+    # HyDE for better semantic matching
+    result = agent.rewrite("What is quantum computing?", strategy=RewriteStrategy.HYDE)
+    print(result.hypothetical_document)
+    
+    # Contextual with chat history
+    chat_history = [
+        {"role": "user", "content": "Tell me about Python"},
+        {"role": "assistant", "content": "Python is a programming language..."}
+    ]
+    result = agent.rewrite("What about its performance?", chat_history=chat_history)
+    
+    # With search tools for context-aware rewriting
+    from praisonaiagents.tools import internet_search
+    agent = QueryRewriterAgent(tools=[internet_search], verbose=True)
+    result = agent.rewrite("latest AI developments")  # Searches first, then rewrites
+"""
+
+import logging
+import json
+from typing import List, Optional, Any, Dict
+from dataclasses import dataclass, field
+from enum import Enum
+
+
+class RewriteStrategy(Enum):
+    """Enumeration of available query rewriting strategies."""
+    BASIC = "basic"
+    HYDE = "hyde"
+    STEP_BACK = "step_back"
+    SUB_QUERIES = "sub_queries"
+    MULTI_QUERY = "multi_query"
+    CONTEXTUAL = "contextual"
+    AUTO = "auto"
+
+
+@dataclass
+class RewriteResult:
+    """Result of a query rewriting operation."""
+    original_query: str
+    rewritten_queries: List[str]
+    strategy_used: RewriteStrategy
+    hypothetical_document: Optional[str] = None
+    step_back_question: Optional[str] = None
+    sub_queries: Optional[List[str]] = None
+    metadata: Dict[str, Any] = field(default_factory=dict)
+    
+    def __repr__(self):
+        return f"RewriteResult(strategy={self.strategy_used.value}, queries={len(self.rewritten_queries)})"
+    
+    @property
+    def primary_query(self) -> str:
+        """Returns the primary rewritten query."""
+        return self.rewritten_queries[0] if self.rewritten_queries else self.original_query
+    
+    @property
+    def all_queries(self) -> List[str]:
+        """Returns all queries including original and rewritten."""
+        queries = [self.original_query]
+        queries.extend(self.rewritten_queries)
+        if self.step_back_question:
+            queries.append(self.step_back_question)
+        if self.sub_queries:
+            queries.extend(self.sub_queries)
+        return list(set(queries))  # Remove duplicates
+
+
+class QueryRewriterAgent:
+    """
+    Agent for rewriting queries to improve retrieval quality in RAG applications.
+    
+    This agent transforms user queries using various strategies to bridge the gap
+    between how users ask questions and how information is stored in knowledge bases.
+    
+    Attributes:
+        name: Name of the agent
+        model: LLM model to use for rewriting
+        verbose: Whether to print detailed logs
+        max_queries: Maximum number of queries to generate for multi-query strategy
+        abbreviations: Dictionary of common abbreviations to expand
+        
+    Example:
+        agent = QueryRewriterAgent(model="gpt-4o-mini", verbose=True)
+        result = agent.rewrite("ML best practices", strategy=RewriteStrategy.MULTI_QUERY)
+        for query in result.rewritten_queries:
+            print(query)
+    """
+    
+    # Default prompts for each strategy
+    PROMPTS = {
+        "basic": """You are an expert at improving prompts and queries.
+
+Given the following input, improve it while preserving its original intent and type.
+
+Guidelines:
+- If it's a TASK (e.g., "write X", "create Y", "generate Z"), keep it as a task - do NOT convert to a question
+- If it's a QUESTION (e.g., "what is X?", "how does Y work?"), keep it as a question
+- Expand abbreviations and acronyms
+- Fix typos and grammatical errors
+- Add relevant context if too short
+- Keep the original intent and action intact
+
+IMPORTANT: If the input is a command/task to DO something, the output must also be a command/task to DO that thing.
+
+Input: {query}
+
+Improved version:""",
+
+        "hyde": """You are an expert at generating hypothetical documents.
+
+Given the following query, write a detailed hypothetical document that would perfectly answer this query. This document will be used to find similar real documents in a knowledge base.
+
+Guidelines:
+- Write as if you are answering the query directly
+- Include specific details, facts, and terminology
+- Use the same style and vocabulary as the target documents
+- Be comprehensive but focused on the query topic
+
+Query: {query}
+
+Hypothetical document:""",
+
+        "step_back": """You are an expert at generating step-back questions.
+
+Given the following query, generate a more general, higher-level question that captures the broader concept. This helps retrieve background information needed to answer the specific query.
+
+Guidelines:
+- Identify the underlying concept or principle
+- Create a question about fundamentals or general knowledge
+- The step-back question should provide context for the original query
+
+Query: {query}
+
+Step-back question:""",
+
+        "sub_queries": """You are an expert at decomposing complex queries.
+
+Given the following query, break it down into simpler, focused sub-queries. Each sub-query should address one specific aspect of the original question.
+
+Guidelines:
+- Identify distinct aspects or components of the query
+- Create independent, self-contained sub-queries
+- Each sub-query should be answerable on its own
+- Return as a JSON array of strings
+
+Query: {query}
+
+Sub-queries (JSON array):""",
+
+        "multi_query": """You are an expert at query expansion and paraphrasing.
+
+Given the following query, generate {num_queries} different versions that capture the same intent but use different words, phrases, or perspectives.
+
+Guidelines:
+- Use synonyms and alternative phrasings
+- Try different question formats (what, how, why, etc.)
+- Include both formal and informal versions
+- Add relevant context or specificity
+- Return as a JSON array of strings
+
+Query: {query}
+
+Alternative queries (JSON array):""",
+
+        "contextual": """You are an expert at understanding conversational context.
+
+Given the conversation history and a follow-up query, rewrite the query as a standalone question that captures all relevant context.
+
+Guidelines:
+- Resolve pronouns and references (it, this, that, etc.)
+- Include relevant context from the conversation
+- Keep entity names and important details
+- Make the query self-contained and clear
+
+Conversation history:
+{chat_history}
+
+Follow-up query: {query}
+
+Standalone question:""",
+
+        "auto": """You are an expert at analyzing and improving prompts and queries.
+
+Analyze the following input and improve it while preserving its original intent and type.
+
+Input types to consider:
+1. TASK/COMMAND (write, create, generate, build) → Keep as a task, make it clearer
+2. QUESTION (what, how, why, when) → Keep as a question, make it more specific
+3. Short/keyword inputs → Expand with context but preserve type
+4. Complex multi-part inputs → Clarify structure
+5. Well-formed inputs → Minor optimization only
+
+IMPORTANT: Never convert a task into a question. "Write X" should remain "Write X", not "How to write X?"
+
+Input: {query}
+{context}
+
+Improved version:"""
+    }
+    
+    def __init__(
+        self,
+        name: str = "QueryRewriterAgent",
+        model: str = "gpt-4o-mini",
+        instructions: Optional[str] = None,
+        verbose: bool = False,
+        max_queries: int = 5,
+        abbreviations: Optional[Dict[str, str]] = None,
+        temperature: float = 0.3,
+        max_tokens: int = 500,
+        tools: Optional[List[Any]] = None
+    ):
+        """
+        Initialize the QueryRewriterAgent.
+        
+        Args:
+            name: Name of the agent
+            model: LLM model to use (default: gpt-4o-mini)
+            instructions: Custom instructions for the agent
+            verbose: Whether to print detailed logs
+            max_queries: Maximum queries for multi-query strategy
+            abbreviations: Custom abbreviation expansions
+            temperature: Temperature for LLM generation
+            max_tokens: Maximum tokens for LLM response
+            tools: Optional list of tools (agent decides when to use them)
+        """
+        self.name = name
+        self.model = model
+        self.instructions = instructions
+        self.verbose = verbose
+        self.max_queries = max_queries
+        self.temperature = temperature
+        self.max_tokens = max_tokens
+        self.tools = tools or []
+        self._agent = None  # Lazy initialized
+        
+        # Default abbreviations (can be extended)
+        self.abbreviations = abbreviations or {
+            "AI": "Artificial Intelligence",
+            "ML": "Machine Learning",
+            "NLP": "Natural Language Processing",
+            "LLM": "Large Language Model",
+            "RAG": "Retrieval Augmented Generation",
+            "API": "Application Programming Interface",
+            "DB": "Database",
+            "UI": "User Interface",
+            "UX": "User Experience",
+            "SDK": "Software Development Kit",
+            "CLI": "Command Line Interface",
+        }
+        
+        # Setup logging
+        self.logger = logging.getLogger(__name__)
+        if self.verbose:
+            self.logger.setLevel(logging.DEBUG)
+    
+    @property
+    def agent(self):
+        """Lazy initialization of internal Agent."""
+        if self._agent is None:
+            from .agent import Agent
+            self._agent = Agent(
+                name=self.name,
+                role="Query Rewriting Specialist",
+                goal="Rewrite queries for better search and retrieval results. Use available tools when needed to gather context.",
+                backstory="You are an expert at understanding user intent and transforming queries for optimal information retrieval.",
+                tools=self.tools,
+                llm=self.model,
+                verbose=self.verbose,
+                markdown=False
+            )
+        return self._agent
+    
+    def _call_agent(self, prompt: str) -> str:
+        """Call the internal Agent with the given prompt."""
+        try:
+            response = self.agent.chat(prompt)
+            return response if response else ""
+        except Exception as e:
+            self.logger.error(f"Agent call failed: {e}")
+            raise
+    
+    def _expand_abbreviations(self, query: str) -> str:
+        """Expand known abbreviations in the query."""
+        words = query.split()
+        expanded = []
+        for word in words:
+            clean_word = word.strip(".,!?;:")
+            if clean_word.upper() in self.abbreviations:
+                expanded.append(f"{clean_word} ({self.abbreviations[clean_word.upper()]})")
+            else:
+                expanded.append(word)
+        return " ".join(expanded)
+    
+    def _is_short_query(self, query: str, threshold: int = 5) -> bool:
+        """Check if query is considered short (keyword-style)."""
+        return len(query.split()) < threshold
+    
+    def _parse_json_array(self, text: str) -> List[str]:
+        """Parse a JSON array from LLM response."""
+        try:
+            # Try direct JSON parsing
+            result = json.loads(text)
+            if isinstance(result, list):
+                return [str(item) for item in result]
+        except json.JSONDecodeError:
+            pass
+        
+        # Try to extract JSON array from text
+        import re
+        match = re.search(r'\[.*?\]', text, re.DOTALL)
+        if match:
+            try:
+                result = json.loads(match.group())
+                if isinstance(result, list):
+                    return [str(item) for item in result]
+            except json.JSONDecodeError:
+                pass
+        
+        # Fallback: split by newlines and clean up
+        lines = [line.strip() for line in text.split('\n') if line.strip()]
+        lines = [line.lstrip('0123456789.-) ').strip('"\'') for line in lines]
+        return [line for line in lines if line and len(line) > 3]
+    
+    def rewrite(
+        self,
+        query: str,
+        strategy: RewriteStrategy = RewriteStrategy.AUTO,
+        chat_history: Optional[List[Dict[str, str]]] = None,
+        context: Optional[str] = None,
+        num_queries: int = None
+    ) -> RewriteResult:
+        """
+        Rewrite a query using the specified strategy.
+        
+        All operations go through an internal Agent which automatically handles
+        tool calling when tools are provided. The agent decides when to use tools
+        based on the query context.
+        
+        Args:
+            query: The original user query
+            strategy: Rewriting strategy to use
+            chat_history: Previous conversation messages for contextual rewriting
+            context: Additional context about the knowledge base
+            num_queries: Number of queries for multi-query strategy
+            
+        Returns:
+            RewriteResult containing rewritten queries
+        """
+        if self.verbose:
+            self.logger.debug(f"Rewriting query: '{query}' with strategy: {strategy.value}")
+            if self.tools:
+                print(f"[cyan]Agent has {len(self.tools)} tools available (will use if needed)...[/cyan]")
+        
+        # Auto-detect strategy if needed
+        if strategy == RewriteStrategy.AUTO:
+            strategy = self._detect_strategy(query, chat_history)
+            if self.verbose:
+                self.logger.debug(f"Auto-detected strategy: {strategy.value}")
+        
+        # All calls go through the internal Agent which handles tool calling
+        combined_context = context or ""
+        
+        # Dispatch to appropriate method
+        if strategy == RewriteStrategy.BASIC:
+            return self._rewrite_basic(query, combined_context)
+        elif strategy == RewriteStrategy.HYDE:
+            return self._rewrite_hyde(query, combined_context)
+        elif strategy == RewriteStrategy.STEP_BACK:
+            return self._rewrite_step_back(query, combined_context)
+        elif strategy == RewriteStrategy.SUB_QUERIES:
+            return self._rewrite_sub_queries(query, combined_context)
+        elif strategy == RewriteStrategy.MULTI_QUERY:
+            return self._rewrite_multi_query(query, num_queries or self.max_queries, combined_context)
+        elif strategy == RewriteStrategy.CONTEXTUAL:
+            return self._rewrite_contextual(query, chat_history or [], combined_context)
+        else:
+            # Default to basic
+            return self._rewrite_basic(query, combined_context)
+    
+    def _detect_strategy(
+        self,
+        query: str,
+        chat_history: Optional[List[Dict[str, str]]] = None
+    ) -> RewriteStrategy:
+        """Auto-detect the best rewriting strategy for a query."""
+        # If chat history exists and query seems like a follow-up
+        if chat_history and self._is_follow_up(query):
+            return RewriteStrategy.CONTEXTUAL
+        
+        # Short keyword queries
+        if self._is_short_query(query):
+            return RewriteStrategy.BASIC
+        
+        # Complex multi-part queries
+        if self._is_complex_query(query):
+            return RewriteStrategy.SUB_QUERIES
+        
+        # Questions that might benefit from step-back
+        if self._needs_step_back(query):
+            return RewriteStrategy.STEP_BACK
+        
+        # Default to basic
+        return RewriteStrategy.BASIC
+    
+    def _is_follow_up(self, query: str) -> bool:
+        """Check if query appears to be a follow-up question."""
+        follow_up_indicators = [
+            "it", "this", "that", "they", "them", "their",
+            "what about", "how about", "and", "also", "more",
+            "same", "similar", "related"
+        ]
+        query_lower = query.lower()
+        return any(indicator in query_lower for indicator in follow_up_indicators)
+    
+    def _is_complex_query(self, query: str) -> bool:
+        """Check if query contains multiple distinct questions."""
+        # Check for multiple question marks or conjunctions
+        if query.count('?') > 1:
+            return True
+        complex_indicators = [" and ", " or ", " also ", " as well as ", " plus "]
+        return any(ind in query.lower() for ind in complex_indicators)
+    
+    def _needs_step_back(self, query: str) -> bool:
+        """Check if query would benefit from a step-back question."""
+        # Queries about specific details often benefit from broader context
+        specific_indicators = [
+            "specific", "exactly", "precisely", "particular",
+            "difference between", "compare", "versus", "vs"
+        ]
+        return any(ind in query.lower() for ind in specific_indicators)
+    
+    def _rewrite_basic(self, query: str, context: str = "") -> RewriteResult:
+        """Basic query rewriting for clarity and keyword optimization."""
+        # First expand abbreviations
+        expanded_query = self._expand_abbreviations(query)
+        
+        # Build prompt with optional context
+        prompt = self.PROMPTS["basic"].format(query=expanded_query)
+        if context:
+            prompt += f"\n\nAdditional context to help with rewriting:\n{context}"
+        
+        rewritten = self._call_agent(prompt).strip()
+        
+        return RewriteResult(
+            original_query=query,
+            rewritten_queries=[rewritten],
+            strategy_used=RewriteStrategy.BASIC,
+            metadata={"expanded_abbreviations": expanded_query != query, "has_context": bool(context)}
+        )
+    
+    def _rewrite_hyde(self, query: str, context: str = "") -> RewriteResult:
+        """HyDE: Generate hypothetical document for better semantic matching."""
+        prompt = self.PROMPTS["hyde"].format(query=query)
+        if context:
+            prompt += f"\n\nUse this context to make the hypothetical document more accurate:\n{context}"
+        
+        hypothetical_doc = self._call_agent(prompt).strip()
+        
+        # The hypothetical document itself becomes the search query
+        return RewriteResult(
+            original_query=query,
+            rewritten_queries=[hypothetical_doc[:500]],  # Truncate for embedding
+            strategy_used=RewriteStrategy.HYDE,
+            hypothetical_document=hypothetical_doc,
+            metadata={"full_document_length": len(hypothetical_doc), "has_context": bool(context)}
+        )
+    
+    def _rewrite_step_back(self, query: str, context: str = "") -> RewriteResult:
+        """Generate step-back question for broader context retrieval."""
+        # Generate step-back question
+        prompt = self.PROMPTS["step_back"].format(query=query)
+        if context:
+            prompt += f"\n\nContext:\n{context}"
+        step_back = self._call_agent(prompt).strip()
+        
+        # Also do basic rewriting of original
+        basic_result = self._rewrite_basic(query, context)
+        
+        return RewriteResult(
+            original_query=query,
+            rewritten_queries=basic_result.rewritten_queries,
+            strategy_used=RewriteStrategy.STEP_BACK,
+            step_back_question=step_back,
+            metadata={"includes_step_back": True, "has_context": bool(context)}
+        )
+    
+    def _rewrite_sub_queries(self, query: str, context: str = "") -> RewriteResult:
+        """Decompose complex query into focused sub-queries."""
+        prompt = self.PROMPTS["sub_queries"].format(query=query)
+        if context:
+            prompt += f"\n\nContext to help with decomposition:\n{context}"
+        response = self._call_agent(prompt).strip()
+        
+        sub_queries = self._parse_json_array(response)
+        
+        # Ensure we have at least one sub-query
+        if not sub_queries:
+            sub_queries = [query]
+        
+        return RewriteResult(
+            original_query=query,
+            rewritten_queries=sub_queries,
+            strategy_used=RewriteStrategy.SUB_QUERIES,
+            sub_queries=sub_queries,
+            metadata={"num_sub_queries": len(sub_queries), "has_context": bool(context)}
+        )
+    
+    def _rewrite_multi_query(self, query: str, num_queries: int, context: str = "") -> RewriteResult:
+        """Generate multiple paraphrased versions for ensemble retrieval."""
+        prompt = self.PROMPTS["multi_query"].format(
+            query=query,
+            num_queries=num_queries
+        )
+        if context:
+            prompt += f"\n\nContext to help with paraphrasing:\n{context}"
+        response = self._call_agent(prompt).strip()
+        
+        queries = self._parse_json_array(response)
+        
+        # Ensure we have queries and limit to max
+        if not queries:
+            queries = [query]
+        queries = queries[:num_queries]
+        
+        return RewriteResult(
+            original_query=query,
+            rewritten_queries=queries,
+            strategy_used=RewriteStrategy.MULTI_QUERY,
+            metadata={"num_generated": len(queries), "has_context": bool(context)}
+        )
+    
+    def _rewrite_contextual(
+        self,
+        query: str,
+        chat_history: List[Dict[str, str]],
+        context: str = ""
+    ) -> RewriteResult:
+        """Rewrite query using conversation context."""
+        # Format chat history
+        history_str = ""
+        for msg in chat_history[-10:]:  # Last 10 messages
+            role = msg.get("role", "user")
+            content = msg.get("content", "")
+            history_str += f"{role.capitalize()}: {content}\n"
+        
+        prompt = self.PROMPTS["contextual"].format(
+            chat_history=history_str,
+            query=query
+        )
+        if context:
+            prompt += f"\n\nAdditional context:\n{context}"
+        rewritten = self._call_agent(prompt).strip()
+        
+        return RewriteResult(
+            original_query=query,
+            rewritten_queries=[rewritten],
+            strategy_used=RewriteStrategy.CONTEXTUAL,
+            metadata={
+                "history_length": len(chat_history),
+                "history_used": min(10, len(chat_history)),
+                "has_context": bool(context)
+            }
+        )
+    
+    # Convenience methods for direct strategy access
+    def rewrite_basic(self, query: str) -> RewriteResult:
+        """Convenience method for basic rewriting."""
+        return self.rewrite(query, strategy=RewriteStrategy.BASIC)
+    
+    def rewrite_hyde(self, query: str) -> RewriteResult:
+        """Convenience method for HyDE rewriting."""
+        return self.rewrite(query, strategy=RewriteStrategy.HYDE)
+    
+    def rewrite_step_back(self, query: str) -> RewriteResult:
+        """Convenience method for step-back rewriting."""
+        return self.rewrite(query, strategy=RewriteStrategy.STEP_BACK)
+    
+    def rewrite_sub_queries(self, query: str) -> RewriteResult:
+        """Convenience method for sub-query decomposition."""
+        return self.rewrite(query, strategy=RewriteStrategy.SUB_QUERIES)
+    
+    def rewrite_multi_query(self, query: str, num_queries: int = None) -> RewriteResult:
+        """Convenience method for multi-query generation."""
+        return self.rewrite(
+            query,
+            strategy=RewriteStrategy.MULTI_QUERY,
+            num_queries=num_queries
+        )
+    
+    def rewrite_contextual(
+        self,
+        query: str,
+        chat_history: List[Dict[str, str]]
+    ) -> RewriteResult:
+        """Convenience method for contextual rewriting."""
+        return self.rewrite(
+            query,
+            strategy=RewriteStrategy.CONTEXTUAL,
+            chat_history=chat_history
+        )
+    
+    def add_abbreviation(self, abbrev: str, expansion: str) -> None:
+        """Add a custom abbreviation expansion."""
+        self.abbreviations[abbrev.upper()] = expansion
+    
+    def add_abbreviations(self, abbreviations: Dict[str, str]) -> None:
+        """Add multiple custom abbreviation expansions."""
+        for abbrev, expansion in abbreviations.items():
+            self.abbreviations[abbrev.upper()] = expansion

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