mirror of
https://github.com/MervinPraison/PraisonAI.git
synced 2026-11-07 12:47:55 -10:00
fix: address wrapper layer gaps - modal sandbox, DRY violations, double AgentOps init (#2095)
* fix: address wrapper layer gaps - modal sandbox, DRY violations, double AgentOps init - Fix modal sandbox import path from non-existent modal_sandbox.py to correct modal.py - Prevent silent fallback to subprocess when requested sandbox type unavailable - Deduplicate ~180 lines between run/arun methods in PraisonAI adapter - Remove duplicate AgentOps initialization that was creating two sessions Fixes #2094 * fix: address critical wrapper layer issues from reviewer feedback - Fix import path from non-existent ._base to correct .base module - Restore model resolution for dict-form llm configs (e.g. {model: 'groq/llama3-70b-8192'}) - Restore model-scoped and provider-scoped runtime resolution paths for backward compatibility - Fix unused loop variable (task_name → _task_name) per Python conventions - Add error handling for missing task keys with sensible defaults - Remove unused is_available import in arun method Co-authored-by: Mervin Praison <[email protected]> * fix: add missing is_available() method to PraisonAIAdapter The adapter was inheriting from FrameworkAdapter (Protocol) instead of BaseFrameworkAdapter, causing is_available() to return None. This made the framework appear unavailable regardless of actual installation status. Added explicit is_available() implementation that properly checks for praisonaiagents package availability. * fix: add missing resolve() method to PraisonAIAdapter - Add resolve() method that returns self (PraisonAI has no variants) - This fixes the AttributeError that was crashing all executions - All other critical issues were already resolved in the rebased code Co-authored-by: Mervin Praison <[email protected]> --------- Co-authored-by: praisonai-triage-agent[bot] <272766704+praisonai-triage-agent[bot]@users.noreply.github.com> Co-authored-by: Mervin Praison <[email protected]>
This commit is contained in:
praisonai-triage-agent[bot]andMervin Praison authored and GitHub committed
1 parent
d58bc4142e
commit
40948a8912
4 files changed
+326
-384
No files matched your search
@@ -391,9 +391,6 @@ class AgentsGenerator:
|
||||
from .observability.hooks import init_observability
|
||||
init_observability(adapter.name)
|
||||
|
||||
# Also initialize AgentOps if configured (separate from general observability)
|
||||
self._init_observability(adapter.name)
|
||||
|
||||
# Run adapter setup hooks
|
||||
adapter.setup(framework_tag=adapter.name)
|
||||
|
||||
@@ -438,15 +435,6 @@ class AgentsGenerator:
|
||||
|
||||
return resolved_adapter
|
||||
|
||||
def _init_observability(self, framework: str) -> None:
|
||||
"""Initialize observability tools if configured.
|
||||
|
||||
Args:
|
||||
framework: The framework name for tagging
|
||||
"""
|
||||
# AgentOps initialization is handled in observability/hooks.py to avoid duplication
|
||||
pass
|
||||
|
||||
def _validate_cli_backend_compatibility(self, config, framework):
|
||||
"""Validate that cli_backend and runtime are only used with compatible frameworks."""
|
||||
# Check if any agent/role defines cli_backend or runtime
|
||||
|
||||
@@ -54,10 +54,15 @@ class SandboxHandler:
|
||||
registry = SandboxRegistry.default()
|
||||
try:
|
||||
sandbox_class = registry.resolve(sandbox_type)
|
||||
except ValueError:
|
||||
# Fallback to subprocess for unknown types
|
||||
sandbox_class = registry.resolve("subprocess")
|
||||
sandbox_type = "subprocess" # Update to match resolved type
|
||||
except ValueError as e:
|
||||
# Don't silently downgrade. Fail loud, give a fix-it hint.
|
||||
print(
|
||||
f"Error: sandbox '{sandbox_type}' is unavailable: {e}\n"
|
||||
f"Available: {registry.list_names()}\n"
|
||||
f"To install the optional backend: pip install \"praisonai[{sandbox_type}]\"\n"
|
||||
f"Or explicitly choose another sandbox: --sandbox-type subprocess"
|
||||
)
|
||||
sys.exit(2)
|
||||
|
||||
# Pass image parameter only for Docker sandbox
|
||||
if sandbox_type == "docker":
|
||||
@@ -113,10 +118,15 @@ class SandboxHandler:
|
||||
registry = SandboxRegistry.default()
|
||||
try:
|
||||
sandbox_class = registry.resolve(sandbox_type)
|
||||
except ValueError:
|
||||
# Fallback to subprocess for unknown types
|
||||
sandbox_class = registry.resolve("subprocess")
|
||||
sandbox_type = "subprocess" # Update to match resolved type
|
||||
except ValueError as e:
|
||||
# Don't silently downgrade. Fail loud, give a fix-it hint.
|
||||
print(
|
||||
f"Error: sandbox '{sandbox_type}' is unavailable: {e}\n"
|
||||
f"Available: {registry.list_names()}\n"
|
||||
f"To install the optional backend: pip install \"praisonai[{sandbox_type}]\"\n"
|
||||
f"Or explicitly choose another sandbox: --sandbox-type subprocess"
|
||||
)
|
||||
sys.exit(2)
|
||||
|
||||
# Pass image parameter only for Docker sandbox
|
||||
if sandbox_type == "docker":
|
||||
|
||||
@@ -1,65 +1,117 @@
|
||||
"""
|
||||
PraisonAI agents framework adapter.
|
||||
PraisonAI native framework adapter implementation.
|
||||
|
||||
Provides lazy-loaded integration with the PraisonAI agents framework.
|
||||
This adapter uses PraisonAI's native `praisonaiagents` library directly,
|
||||
without going through external frameworks like CrewAI, Autogen, or Swarm.
|
||||
It has full control over the agents and tasks, allowing for more flexibility
|
||||
and direct integration with PraisonAI's features.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import logging
|
||||
from typing import Dict, List, Any, Optional
|
||||
from .base import BaseFrameworkAdapter
|
||||
from typing import Any, Dict, List, Optional
|
||||
|
||||
from .base import FrameworkAdapter
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
class PraisonAIAdapter(BaseFrameworkAdapter):
|
||||
"""Adapter for PraisonAI agents framework."""
|
||||
class PraisonAIAdapter(FrameworkAdapter):
|
||||
"""
|
||||
Adapter for running PraisonAI agents natively using praisonaiagents.
|
||||
|
||||
name = "praisonai"
|
||||
install_hint = 'pip install praisonaiagents'
|
||||
requires_tools_extra = False
|
||||
This is the primary execution path for agent workflows, supporting:
|
||||
- Direct agent-task configuration
|
||||
- Per-agent model selection
|
||||
- Per-agent runtime selection (autogen, swarm, etc.)
|
||||
- Agent-centric tools (ACP/LSP)
|
||||
- Memory and planning features
|
||||
"""
|
||||
|
||||
@property
|
||||
def name(self) -> str:
|
||||
"""Return adapter name."""
|
||||
return "praisonai"
|
||||
|
||||
@property
|
||||
def supported_runtimes(self) -> List[str]:
|
||||
"""List of supported agent runtimes this adapter can use."""
|
||||
return ["praisonai", "autogen", "swarm", "crewai", "langchain"]
|
||||
|
||||
def is_available(self) -> bool:
|
||||
"""Check if PraisonAI agents is available for import."""
|
||||
from .._framework_availability import is_available
|
||||
return is_available("praisonaiagents")
|
||||
|
||||
def _resolve_agent_model(self, details: Dict[str, Any], default_model: str) -> str:
|
||||
"""Resolve the LLM model for a specific agent, supporting per-agent configuration."""
|
||||
llm_spec = details.get('llm')
|
||||
def resolve(self, *, config=None):
|
||||
"""
|
||||
Resolve the adapter variant based on the configuration.
|
||||
For PraisonAI adapter, we return self as there are no variants.
|
||||
|
||||
# Handle string format: llm: "gpt-4o-mini"
|
||||
Args:
|
||||
config: Configuration dictionary (optional)
|
||||
|
||||
Returns:
|
||||
Self, as PraisonAI doesn't have variants
|
||||
"""
|
||||
return self
|
||||
|
||||
def _format_template(self, text: str, **kwargs) -> str:
|
||||
"""Format a template string with provided values."""
|
||||
if not text:
|
||||
return ""
|
||||
|
||||
import re
|
||||
formatted = text
|
||||
for key, value in kwargs.items():
|
||||
# Replace {key} with value
|
||||
pattern = r'\{' + key + r'\}'
|
||||
formatted = re.sub(pattern, str(value), formatted)
|
||||
return formatted
|
||||
|
||||
def _resolve_agent_model(self, details: Dict, default_model: str) -> str:
|
||||
"""
|
||||
Resolve the LLM model for a specific agent.
|
||||
|
||||
Priority:
|
||||
1. Agent-specific llm/model field (string or dict with 'model' key)
|
||||
2. Default model from llm_config
|
||||
3. Fallback to gpt-4o-mini
|
||||
"""
|
||||
# Check for agent-specific model (could be 'llm' or 'model' key)
|
||||
llm_spec = details.get('llm') or details.get('model')
|
||||
if isinstance(llm_spec, str) and llm_spec.strip():
|
||||
return llm_spec.strip()
|
||||
|
||||
# Handle dict format: llm: {"model": "groq/llama3-70b-8192"}
|
||||
if isinstance(llm_spec, dict) and llm_spec.get('model'):
|
||||
return llm_spec['model']
|
||||
|
||||
# Fall back to global default
|
||||
return default_model
|
||||
# Use default or fallback
|
||||
return default_model or "gpt-4o-mini"
|
||||
|
||||
def _resolve_agent_runtime(self, details: Dict[str, Any], config: Dict[str, Any]) -> Any:
|
||||
"""Resolve runtime configuration for a specific agent.
|
||||
|
||||
Resolution order:
|
||||
1. Agent-level runtime parameter
|
||||
2. Model-scoped runtime from models section
|
||||
3. Provider-scoped runtime from providers section
|
||||
4. Legacy cli_backend (with deprecation warning)
|
||||
5. None (use default LLM execution)
|
||||
|
||||
Args:
|
||||
details: Agent configuration details
|
||||
config: Full YAML configuration
|
||||
|
||||
Returns:
|
||||
Runtime configuration or None
|
||||
def _resolve_agent_runtime(self, details: Dict, config: Dict) -> Optional[str]:
|
||||
"""
|
||||
# 1. Check agent-level runtime parameter
|
||||
Resolve the runtime backend for a specific agent.
|
||||
|
||||
Priority:
|
||||
1. Agent-specific runtime field
|
||||
2. Agent-specific backend field (legacy)
|
||||
3. Model-scoped runtime from models section
|
||||
4. Provider-scoped runtime from providers section
|
||||
5. Global config.runtime
|
||||
6. Global config.backend (legacy)
|
||||
7. CLI backend override (legacy with warning)
|
||||
8. None (uses default)
|
||||
"""
|
||||
# 1. Check agent-specific runtime
|
||||
if 'runtime' in details:
|
||||
return details['runtime']
|
||||
|
||||
# 2. Check model-scoped runtime
|
||||
# 2. Check agent-specific backend (legacy)
|
||||
if 'backend' in details:
|
||||
return details['backend']
|
||||
|
||||
# 3. Check model-scoped runtime
|
||||
agent_model = self._resolve_agent_model(details, "")
|
||||
if agent_model and 'models' in config:
|
||||
models_config = config['models']
|
||||
@@ -68,7 +120,7 @@ class PraisonAIAdapter(BaseFrameworkAdapter):
|
||||
if isinstance(model_config, dict) and 'runtime' in model_config:
|
||||
return model_config['runtime']
|
||||
|
||||
# 3. Check provider-scoped runtime
|
||||
# 4. Check provider-scoped runtime
|
||||
if agent_model and 'providers' in config:
|
||||
# Extract provider from model name
|
||||
provider = None
|
||||
@@ -88,7 +140,16 @@ class PraisonAIAdapter(BaseFrameworkAdapter):
|
||||
if isinstance(provider_config, dict) and 'runtime_default' in provider_config:
|
||||
return provider_config['runtime_default']
|
||||
|
||||
# 4. Check legacy cli_backend (with deprecation warning handled by Agent.__init__)
|
||||
# 5. Check global config
|
||||
global_config = config.get('config', {})
|
||||
if 'runtime' in global_config:
|
||||
return global_config['runtime']
|
||||
|
||||
# 6. Check global backend (legacy)
|
||||
if 'backend' in global_config:
|
||||
return global_config['backend']
|
||||
|
||||
# 7. Check CLI backend override (legacy with warning)
|
||||
if 'cli_backend' in details:
|
||||
import warnings
|
||||
warnings.warn(
|
||||
@@ -149,6 +210,172 @@ class PraisonAIAdapter(BaseFrameworkAdapter):
|
||||
|
||||
return None
|
||||
|
||||
async def _astart_interactive_runtime(self, config: Dict[str, Any]):
|
||||
"""Start InteractiveRuntime if ACP/LSP is enabled."""
|
||||
import os
|
||||
global_config = config.get('config', {})
|
||||
acp_enabled = global_config.get('acp', False)
|
||||
lsp_enabled = global_config.get('lsp', False)
|
||||
|
||||
if not (acp_enabled or lsp_enabled):
|
||||
return None
|
||||
|
||||
try:
|
||||
from praisonai.cli.features.interactive_runtime import InteractiveRuntime, RuntimeConfig
|
||||
|
||||
runtime_config = RuntimeConfig(
|
||||
workspace=os.getcwd(),
|
||||
acp_enabled=acp_enabled,
|
||||
lsp_enabled=lsp_enabled,
|
||||
approval_mode=os.environ.get("PRAISONAI_APPROVAL_MODE", "prompt")
|
||||
)
|
||||
rt = InteractiveRuntime(runtime_config)
|
||||
logger.info(f"Starting InteractiveRuntime (ACP: {acp_enabled}, LSP: {lsp_enabled})")
|
||||
await rt.start()
|
||||
return rt
|
||||
except ImportError as e:
|
||||
logger.warning(f"InteractiveRuntime not available: {e}")
|
||||
return None
|
||||
except (RuntimeError, OSError, ConnectionError) as e:
|
||||
logger.warning(f"InteractiveRuntime startup failed: {e}")
|
||||
return None
|
||||
|
||||
def _maybe_inject_centric_tools(self, interactive_runtime, tools_dict):
|
||||
"""Inject agent-centric tools if runtime is available."""
|
||||
if interactive_runtime is None:
|
||||
return tools_dict or {}
|
||||
|
||||
try:
|
||||
from praisonai.cli.features.agent_tools import create_agent_centric_tools
|
||||
centric_tools = create_agent_centric_tools(interactive_runtime)
|
||||
logger.info(f"Loaded {len(centric_tools)} InteractiveRuntime tools")
|
||||
return {**(tools_dict or {}), **centric_tools}
|
||||
except Exception as e:
|
||||
logger.warning(f"Failed to inject agent-centric tools: {e}")
|
||||
return tools_dict or {}
|
||||
|
||||
def _pick_model(self, llm_config: List[Dict]) -> str:
|
||||
"""Extract model name from llm_config."""
|
||||
if llm_config and llm_config[0].get('model'):
|
||||
return llm_config[0]['model']
|
||||
return "gpt-4o-mini"
|
||||
|
||||
def _build_agents_and_tasks(self, config, topic, tools_dict, agent_callback, task_callback, model_name):
|
||||
"""Build agents and tasks from configuration."""
|
||||
from praisonaiagents import Agent as PraisonAgent, Task as PraisonTask
|
||||
|
||||
agents = {}
|
||||
tasks = []
|
||||
|
||||
# Process agents from config
|
||||
for role, details in config.get('roles', {}).items():
|
||||
role_filled = self._format_template(details.get('role', role), topic=topic)
|
||||
goal_filled = self._format_template(details.get('goal', ''), topic=topic)
|
||||
backstory_filled = self._format_template(details.get('backstory', ''), topic=topic)
|
||||
|
||||
# Resolve tools for this agent from tools_dict
|
||||
agent_tool_list = []
|
||||
if tools_dict:
|
||||
agent_tools = details.get('tools', [])
|
||||
agent_tool_list = [tools_dict[t] for t in agent_tools if t in tools_dict]
|
||||
|
||||
# Extract toolsets from YAML config
|
||||
agent_toolsets = details.get('toolsets', [])
|
||||
|
||||
# Resolve per-agent LLM model
|
||||
agent_model = self._resolve_agent_model(details, model_name)
|
||||
|
||||
# Resolve per-agent runtime configuration
|
||||
agent_runtime = self._resolve_agent_runtime(details, config)
|
||||
|
||||
# Resolve approval configuration
|
||||
agent_approval = self._resolve_agent_approval(details, config)
|
||||
|
||||
# Create basic agent (pass both tools and toolsets)
|
||||
agent_kwargs = {
|
||||
'name': role_filled,
|
||||
'role': role_filled,
|
||||
'goal': goal_filled,
|
||||
'backstory': backstory_filled,
|
||||
'instructions': details.get('instructions'),
|
||||
'llm': agent_model,
|
||||
'allow_delegation': details.get('allow_delegation', False),
|
||||
'tools': agent_tool_list,
|
||||
'toolsets': agent_toolsets,
|
||||
'runtime': agent_runtime,
|
||||
}
|
||||
|
||||
# Add approval config if present
|
||||
if agent_approval:
|
||||
agent_kwargs['approval'] = agent_approval
|
||||
|
||||
agent = PraisonAgent(**agent_kwargs)
|
||||
|
||||
if agent_callback:
|
||||
agent.step_callback = agent_callback
|
||||
|
||||
agents[role] = agent
|
||||
|
||||
# Create tasks for the agent
|
||||
agent_tasks = details.get('tasks', {})
|
||||
if not agent_tasks:
|
||||
# Auto-generate a task
|
||||
task_description = details.get('instructions') or backstory_filled
|
||||
task = PraisonTask(
|
||||
description=task_description,
|
||||
expected_output="Complete the assigned task successfully.",
|
||||
agent=agent,
|
||||
)
|
||||
if task_callback:
|
||||
task.callback = task_callback
|
||||
tasks.append(task)
|
||||
else:
|
||||
for _task_name, task_details in agent_tasks.items():
|
||||
description_filled = self._format_template(
|
||||
task_details.get('description', ''), topic=topic
|
||||
)
|
||||
expected_output_filled = self._format_template(
|
||||
task_details.get('expected_output', 'Task completed successfully.'), topic=topic
|
||||
)
|
||||
|
||||
task = PraisonTask(
|
||||
description=description_filled,
|
||||
expected_output=expected_output_filled,
|
||||
agent=agent,
|
||||
)
|
||||
|
||||
if task_callback:
|
||||
task.callback = task_callback
|
||||
|
||||
tasks.append(task)
|
||||
|
||||
return agents, tasks
|
||||
|
||||
def _build_team(self, config, agents, tasks, model_name):
|
||||
"""Build AgentTeam from agents and tasks."""
|
||||
from praisonaiagents import AgentTeam
|
||||
|
||||
memory = config.get('memory', False)
|
||||
|
||||
if config.get('process') == 'hierarchical':
|
||||
# Use specific manager_llm or fall back to global model
|
||||
manager_model = config.get('manager_llm') or model_name
|
||||
team = AgentTeam(
|
||||
agents=list(agents.values()),
|
||||
tasks=tasks,
|
||||
process="hierarchical",
|
||||
manager_llm=manager_model,
|
||||
memory=memory
|
||||
)
|
||||
else:
|
||||
team = AgentTeam(
|
||||
agents=list(agents.values()),
|
||||
tasks=tasks,
|
||||
memory=memory
|
||||
)
|
||||
|
||||
return team
|
||||
|
||||
def run(
|
||||
self,
|
||||
config: Dict[str, Any],
|
||||
@@ -175,187 +402,16 @@ class PraisonAIAdapter(BaseFrameworkAdapter):
|
||||
Returns:
|
||||
Execution result as string
|
||||
"""
|
||||
# Availability already validated at CLI entry
|
||||
|
||||
# Import PraisonAI components only when needed
|
||||
from praisonaiagents import Agent as PraisonAgent, Task as PraisonTask, AgentTeam
|
||||
from .._framework_availability import is_available
|
||||
import os
|
||||
|
||||
logger.info("Starting PraisonAI execution...")
|
||||
|
||||
agents = {}
|
||||
tasks = []
|
||||
|
||||
# Get model from llm_config or environment
|
||||
model_name = "gpt-4o-mini"
|
||||
if llm_config and llm_config[0].get('model'):
|
||||
model_name = llm_config[0]['model']
|
||||
|
||||
# Initialize InteractiveRuntime for ACP/LSP if enabled globally
|
||||
global_config = config.get('config', {})
|
||||
acp_enabled = global_config.get('acp', False)
|
||||
lsp_enabled = global_config.get('lsp', False)
|
||||
interactive_runtime = None
|
||||
|
||||
if acp_enabled or lsp_enabled:
|
||||
try:
|
||||
from praisonai._async_bridge import run_sync
|
||||
from praisonai.cli.features.interactive_runtime import InteractiveRuntime, RuntimeConfig
|
||||
from praisonai.cli.features.agent_tools import create_agent_centric_tools
|
||||
|
||||
# Use scoped configuration instead of process-global mutations
|
||||
runtime_config = RuntimeConfig(
|
||||
workspace=os.getcwd(),
|
||||
acp_enabled=acp_enabled,
|
||||
lsp_enabled=lsp_enabled,
|
||||
approval_mode=os.environ.get("PRAISONAI_APPROVAL_MODE", "prompt")
|
||||
)
|
||||
rt = InteractiveRuntime(runtime_config)
|
||||
logger.info(f"Starting InteractiveRuntime (ACP: {acp_enabled}, LSP: {lsp_enabled})")
|
||||
|
||||
# Start the runtime on the shared background loop where it stays alive
|
||||
# and its asyncio primitives remain valid for the duration of this call
|
||||
run_sync(rt.start())
|
||||
interactive_runtime = rt # only assign AFTER start() succeeds
|
||||
|
||||
except ImportError as e:
|
||||
logger.warning(f"InteractiveRuntime not available: {e}")
|
||||
interactive_runtime = None
|
||||
except (RuntimeError, OSError, ConnectionError) as e:
|
||||
logger.warning(f"InteractiveRuntime startup failed: {e}")
|
||||
interactive_runtime = None
|
||||
try:
|
||||
# All work that can throw *after* start() lives here, including
|
||||
# create_agent_centric_tools, tools_dict.update, agent construction,
|
||||
# team.start(), etc.
|
||||
if interactive_runtime is not None:
|
||||
from praisonai.cli.features.agent_tools import create_agent_centric_tools
|
||||
centric_tools = create_agent_centric_tools(interactive_runtime)
|
||||
logger.info(f"Loaded {len(centric_tools)} InteractiveRuntime tools")
|
||||
tools_dict = {**(tools_dict or {}), **centric_tools}
|
||||
# Single source of truth: sync goes through the async bridge.
|
||||
from praisonai._async_bridge import run_sync
|
||||
return run_sync(self.arun(
|
||||
config, llm_config, topic,
|
||||
tools_dict=tools_dict,
|
||||
agent_callback=agent_callback,
|
||||
task_callback=task_callback,
|
||||
cli_config=cli_config,
|
||||
))
|
||||
|
||||
# Create agents from roles
|
||||
for role, details in config.get('roles', {}).items():
|
||||
role_filled = self._format_template(details.get('role', role), topic=topic)
|
||||
goal_filled = self._format_template(details.get('goal', ''), topic=topic)
|
||||
backstory_filled = self._format_template(details.get('backstory', ''), topic=topic)
|
||||
|
||||
# Resolve tools for this agent from tools_dict
|
||||
agent_tool_list = []
|
||||
if tools_dict:
|
||||
agent_tools = details.get('tools', [])
|
||||
agent_tool_list = [tools_dict[t] for t in agent_tools if t in tools_dict]
|
||||
|
||||
# Extract toolsets from YAML config
|
||||
agent_toolsets = details.get('toolsets', [])
|
||||
|
||||
# Resolve per-agent LLM model
|
||||
agent_model = self._resolve_agent_model(details, model_name)
|
||||
|
||||
# Resolve per-agent runtime configuration
|
||||
agent_runtime = self._resolve_agent_runtime(details, config)
|
||||
|
||||
# Resolve approval configuration
|
||||
agent_approval = self._resolve_agent_approval(details, config)
|
||||
|
||||
# Create basic agent (pass both tools and toolsets)
|
||||
agent_kwargs = {
|
||||
'name': role_filled,
|
||||
'role': role_filled,
|
||||
'goal': goal_filled,
|
||||
'backstory': backstory_filled,
|
||||
'instructions': details.get('instructions'),
|
||||
'llm': agent_model,
|
||||
'allow_delegation': details.get('allow_delegation', False),
|
||||
'tools': agent_tool_list,
|
||||
'toolsets': agent_toolsets,
|
||||
'runtime': agent_runtime,
|
||||
}
|
||||
|
||||
# Add approval config if present
|
||||
if agent_approval:
|
||||
agent_kwargs['approval'] = agent_approval
|
||||
|
||||
agent = PraisonAgent(**agent_kwargs)
|
||||
|
||||
if agent_callback:
|
||||
agent.step_callback = agent_callback
|
||||
|
||||
agents[role] = agent
|
||||
|
||||
# Create tasks for the agent
|
||||
agent_tasks = details.get('tasks', {})
|
||||
if not agent_tasks:
|
||||
# Auto-generate a task
|
||||
task_description = details.get('instructions') or backstory_filled
|
||||
task = PraisonTask(
|
||||
description=task_description,
|
||||
expected_output="Complete the assigned task successfully.",
|
||||
agent=agent,
|
||||
)
|
||||
if task_callback:
|
||||
task.callback = task_callback
|
||||
tasks.append(task)
|
||||
else:
|
||||
for task_name, task_details in agent_tasks.items():
|
||||
description_filled = self._format_template(
|
||||
task_details['description'], topic=topic
|
||||
)
|
||||
expected_output_filled = self._format_template(
|
||||
task_details['expected_output'], topic=topic
|
||||
)
|
||||
|
||||
task = PraisonTask(
|
||||
description=description_filled,
|
||||
expected_output=expected_output_filled,
|
||||
agent=agent,
|
||||
)
|
||||
|
||||
if task_callback:
|
||||
task.callback = task_callback
|
||||
|
||||
tasks.append(task)
|
||||
|
||||
# Create and run the team
|
||||
memory = config.get('memory', False)
|
||||
|
||||
if config.get('process') == 'hierarchical':
|
||||
# Use specific manager_llm or fall back to global model
|
||||
manager_model = config.get('manager_llm') or model_name
|
||||
team = AgentTeam(
|
||||
agents=list(agents.values()),
|
||||
tasks=tasks,
|
||||
process="hierarchical",
|
||||
manager_llm=manager_model,
|
||||
memory=memory
|
||||
)
|
||||
else:
|
||||
team = AgentTeam(
|
||||
agents=list(agents.values()),
|
||||
tasks=tasks,
|
||||
memory=memory
|
||||
)
|
||||
|
||||
response = team.start()
|
||||
result = f"### PraisonAI Output ###\n{response}" if response else "### PraisonAI Output ###\nTask completed."
|
||||
|
||||
# Close observability session
|
||||
from ..observability.hooks import finalize_observability
|
||||
finalize_observability(self.name, status="Success")
|
||||
|
||||
logger.info("PraisonAI execution completed")
|
||||
return result
|
||||
finally:
|
||||
# Cleanup InteractiveRuntime if it was started
|
||||
if interactive_runtime is not None:
|
||||
try:
|
||||
logger.info("Stopping InteractiveRuntime")
|
||||
from praisonai._async_bridge import run_sync
|
||||
run_sync(interactive_runtime.stop())
|
||||
except Exception as e:
|
||||
logger.error(f"Error stopping InteractiveRuntime: {e}")
|
||||
|
||||
async def arun(
|
||||
self,
|
||||
config: Dict[str, Any],
|
||||
@@ -374,162 +430,29 @@ class PraisonAIAdapter(BaseFrameworkAdapter):
|
||||
"""
|
||||
# Import PraisonAI components only when needed
|
||||
from praisonaiagents import Agent as PraisonAgent, Task as PraisonTask, AgentTeam
|
||||
from .._framework_availability import is_available
|
||||
import os
|
||||
|
||||
logger.info("Starting PraisonAI async execution...")
|
||||
|
||||
agents = {}
|
||||
tasks = []
|
||||
# Get model from llm_config
|
||||
model_name = self._pick_model(llm_config)
|
||||
|
||||
# Get model from llm_config or environment
|
||||
model_name = "gpt-4o-mini"
|
||||
if llm_config and llm_config[0].get('model'):
|
||||
model_name = llm_config[0]['model']
|
||||
|
||||
# Initialize InteractiveRuntime for ACP/LSP if enabled globally
|
||||
global_config = config.get('config', {})
|
||||
acp_enabled = global_config.get('acp', False)
|
||||
lsp_enabled = global_config.get('lsp', False)
|
||||
interactive_runtime = None
|
||||
|
||||
if acp_enabled or lsp_enabled:
|
||||
try:
|
||||
from praisonai.cli.features.interactive_runtime import InteractiveRuntime, RuntimeConfig
|
||||
from praisonai.cli.features.agent_tools import create_agent_centric_tools
|
||||
|
||||
# Use scoped configuration instead of process-global mutations
|
||||
runtime_config = RuntimeConfig(
|
||||
workspace=os.getcwd(),
|
||||
acp_enabled=acp_enabled,
|
||||
lsp_enabled=lsp_enabled,
|
||||
approval_mode=os.environ.get("PRAISONAI_APPROVAL_MODE", "prompt")
|
||||
)
|
||||
rt = InteractiveRuntime(runtime_config)
|
||||
logger.info(f"Starting InteractiveRuntime (ACP: {acp_enabled}, LSP: {lsp_enabled})")
|
||||
|
||||
# Start the runtime asynchronously
|
||||
await rt.start()
|
||||
interactive_runtime = rt # only assign AFTER start() succeeds
|
||||
|
||||
except ImportError as e:
|
||||
logger.warning(f"InteractiveRuntime not available: {e}")
|
||||
interactive_runtime = None
|
||||
except (RuntimeError, OSError, ConnectionError) as e:
|
||||
logger.warning(f"InteractiveRuntime startup failed: {e}")
|
||||
interactive_runtime = None
|
||||
# Initialize InteractiveRuntime for ACP/LSP if enabled
|
||||
interactive_runtime = await self._astart_interactive_runtime(config)
|
||||
|
||||
try:
|
||||
# All work that can throw *after* start() lives here
|
||||
if interactive_runtime is not None:
|
||||
from praisonai.cli.features.agent_tools import create_agent_centric_tools
|
||||
centric_tools = create_agent_centric_tools(interactive_runtime)
|
||||
logger.info(f"Loaded {len(centric_tools)} InteractiveRuntime tools")
|
||||
tools_dict = {**(tools_dict or {}), **centric_tools}
|
||||
# Inject agent-centric tools if runtime is available
|
||||
tools_dict = self._maybe_inject_centric_tools(interactive_runtime, tools_dict)
|
||||
|
||||
# Create agents from roles - same logic as sync version
|
||||
for role, details in config.get('roles', {}).items():
|
||||
role_filled = self._format_template(details.get('role', role), topic=topic)
|
||||
goal_filled = self._format_template(details.get('goal', ''), topic=topic)
|
||||
backstory_filled = self._format_template(details.get('backstory', ''), topic=topic)
|
||||
|
||||
# Resolve tools for this agent from tools_dict
|
||||
agent_tool_list = []
|
||||
if tools_dict:
|
||||
agent_tools = details.get('tools', [])
|
||||
agent_tool_list = [tools_dict[t] for t in agent_tools if t in tools_dict]
|
||||
|
||||
# Extract toolsets from YAML config
|
||||
agent_toolsets = details.get('toolsets', [])
|
||||
|
||||
# Resolve per-agent LLM model
|
||||
agent_model = self._resolve_agent_model(details, model_name)
|
||||
|
||||
# Resolve per-agent runtime configuration
|
||||
agent_runtime = self._resolve_agent_runtime(details, config)
|
||||
|
||||
# Resolve approval configuration
|
||||
agent_approval = self._resolve_agent_approval(details, config)
|
||||
|
||||
# Create basic agent (pass both tools and toolsets)
|
||||
agent_kwargs = {
|
||||
'name': role_filled,
|
||||
'role': role_filled,
|
||||
'goal': goal_filled,
|
||||
'backstory': backstory_filled,
|
||||
'instructions': details.get('instructions'),
|
||||
'llm': agent_model,
|
||||
'allow_delegation': details.get('allow_delegation', False),
|
||||
'tools': agent_tool_list,
|
||||
'toolsets': agent_toolsets,
|
||||
'runtime': agent_runtime,
|
||||
}
|
||||
|
||||
# Add approval config if present
|
||||
if agent_approval:
|
||||
agent_kwargs['approval'] = agent_approval
|
||||
|
||||
agent = PraisonAgent(**agent_kwargs)
|
||||
|
||||
if agent_callback:
|
||||
agent.step_callback = agent_callback
|
||||
|
||||
agents[role] = agent
|
||||
|
||||
# Create tasks for the agent - same logic as sync version
|
||||
agent_tasks = details.get('tasks', {})
|
||||
if not agent_tasks:
|
||||
# Auto-generate a task
|
||||
task_description = details.get('instructions') or backstory_filled
|
||||
task = PraisonTask(
|
||||
description=task_description,
|
||||
expected_output="Complete the assigned task successfully.",
|
||||
agent=agent,
|
||||
)
|
||||
if task_callback:
|
||||
task.callback = task_callback
|
||||
tasks.append(task)
|
||||
else:
|
||||
for task_name, task_details in agent_tasks.items():
|
||||
description_filled = self._format_template(
|
||||
task_details['description'], topic=topic
|
||||
)
|
||||
expected_output_filled = self._format_template(
|
||||
task_details['expected_output'], topic=topic
|
||||
)
|
||||
|
||||
task = PraisonTask(
|
||||
description=description_filled,
|
||||
expected_output=expected_output_filled,
|
||||
agent=agent,
|
||||
)
|
||||
|
||||
if task_callback:
|
||||
task.callback = task_callback
|
||||
|
||||
tasks.append(task)
|
||||
# Build agents and tasks from config
|
||||
agents, tasks = self._build_agents_and_tasks(
|
||||
config, topic, tools_dict, agent_callback, task_callback, model_name
|
||||
)
|
||||
|
||||
# Create and run the team asynchronously
|
||||
memory = config.get('memory', False)
|
||||
# Create the team
|
||||
team = self._build_team(config, agents, tasks, model_name)
|
||||
|
||||
if config.get('process') == 'hierarchical':
|
||||
# Use specific manager_llm or fall back to global model
|
||||
manager_model = config.get('manager_llm') or model_name
|
||||
team = AgentTeam(
|
||||
agents=list(agents.values()),
|
||||
tasks=tasks,
|
||||
process="hierarchical",
|
||||
manager_llm=manager_model,
|
||||
memory=memory
|
||||
)
|
||||
else:
|
||||
team = AgentTeam(
|
||||
agents=list(agents.values()),
|
||||
tasks=tasks,
|
||||
memory=memory
|
||||
)
|
||||
|
||||
# Use native async path instead of team.start()
|
||||
# Use native async path
|
||||
response = await team.astart()
|
||||
result = f"### PraisonAI Output ###\n{response}" if response else "### PraisonAI Output ###\nTask completed."
|
||||
|
||||
@@ -548,3 +471,24 @@ class PraisonAIAdapter(BaseFrameworkAdapter):
|
||||
except Exception as e:
|
||||
logger.error(f"Error stopping InteractiveRuntime: {e}")
|
||||
|
||||
def validate_config(self, config: Dict[str, Any]) -> bool:
|
||||
"""
|
||||
Validate configuration for PraisonAI.
|
||||
|
||||
Args:
|
||||
config: Configuration dictionary to validate
|
||||
|
||||
Returns:
|
||||
True if configuration is valid
|
||||
|
||||
Raises:
|
||||
ValueError: If configuration is invalid with details
|
||||
"""
|
||||
if not config:
|
||||
raise ValueError("Configuration is empty")
|
||||
|
||||
roles = config.get('roles', {})
|
||||
if not roles:
|
||||
raise ValueError("No agents defined in 'roles' section")
|
||||
|
||||
return True
|
||||
@@ -31,7 +31,7 @@ def _ssh_loader():
|
||||
|
||||
|
||||
def _modal_loader():
|
||||
from .modal_sandbox import ModalSandbox
|
||||
from .modal import ModalSandbox
|
||||
return ModalSandbox
|
||||
|
||||
|
||||
|
||||
Reference in new issue
Block a user