jaf-py 2.4.3__py3-none-any.whl → 2.4.5__py3-none-any.whl

jaf/core/__init__.py

@@ -13,6 +13,14 @@ from .agent_tool import (
     get_current_run_config,
     set_current_run_config,
 )
+from .parallel_agents import (
+    ParallelAgentGroup,
+    ParallelExecutionConfig,
+    create_parallel_agents_tool,
+    create_simple_parallel_tool,
+    create_language_specialists_tool,
+    create_domain_experts_tool,
+)
 from .proxy import ProxyConfig, ProxyAuth, create_proxy_config, get_default_proxy_config
 
 __all__ = [
@@ -23,6 +31,8 @@ __all__ = [
     "Message",
     "ModelConfig",
     "ModelProvider",
+    "ParallelAgentGroup",
+    "ParallelExecutionConfig",
     "ProxyAuth",
     "ProxyConfig",
     "RunConfig",
@@ -41,9 +51,13 @@ __all__ = [
     "create_agent_tool",
     "create_conditional_enabler",
     "create_default_output_extractor",
+    "create_domain_experts_tool",
     "create_json_output_extractor",
+    "create_language_specialists_tool",
+    "create_parallel_agents_tool",
     "create_proxy_config",
     "create_run_id",
+    "create_simple_parallel_tool",
     "create_trace_id",
     "get_current_run_config",
     "get_default_proxy_config",

jaf/core/engine.py

@@ -174,16 +174,18 @@ async def run(
         from .agent_tool import set_current_run_config
         set_current_run_config(config)
         
+        state_with_memory = await _load_conversation_history(initial_state, config)
+        
+        # Emit RunStartEvent AFTER loading conversation history so we have complete context
         if config.on_event:
             config.on_event(RunStartEvent(data=to_event_data(RunStartEventData(
                 run_id=initial_state.run_id,
                 trace_id=initial_state.trace_id,
                 session_id=config.conversation_id,
-                context=initial_state.context,
-                messages=initial_state.messages
+                context=state_with_memory.context,
+                messages=state_with_memory.messages,  # Now includes full conversation history
+                agent_name=state_with_memory.current_agent_name
             ))))
-
-        state_with_memory = await _load_conversation_history(initial_state, config)
         
         # Load approvals from storage if configured
         if config.approval_storage:
@@ -514,12 +516,15 @@ async def _run_internal(
                     if len(partial_tool_calls) > 0:
                         message_tool_calls = []
                         for i, tc in enumerate(partial_tool_calls):
+                            arguments = tc["function"]["arguments"]
+                            if isinstance(arguments, str):
+                                arguments = _normalize_tool_call_arguments(arguments)
                             message_tool_calls.append({
                                 "id": tc["id"] or f"call_{i}",
                                 "type": "function",
                                 "function": {
                                     "name": tc["function"]["name"] or "",
-                                    "arguments": tc["function"]["arguments"]
+                                    "arguments": arguments
                                 }
                             })
 
@@ -532,7 +537,7 @@ async def _run_internal(
                                 type="function",
                                 function=ToolCallFunction(
                                     name=mc["function"]["name"],
-                                    arguments=mc["function"]["arguments"],
+                                    arguments=_normalize_tool_call_arguments(mc["function"]["arguments"])
                                 ),
                             ) for mc in message_tool_calls
                         ],
@@ -551,12 +556,15 @@ async def _run_internal(
             if len(partial_tool_calls) > 0:
                 final_tool_calls = []
                 for i, tc in enumerate(partial_tool_calls):
+                    arguments = tc["function"]["arguments"]
+                    if isinstance(arguments, str):
+                        arguments = _normalize_tool_call_arguments(arguments)
                     final_tool_calls.append({
                         "id": tc["id"] or f"call_{i}",
                         "type": "function",
                         "function": {
                             "name": tc["function"]["name"] or "",
-                            "arguments": tc["function"]["arguments"]
+                            "arguments": arguments
                         }
                     })
 
@@ -842,12 +850,33 @@ def _convert_tool_calls(tool_calls: Optional[List[Dict[str, Any]]]) -> Optional[
             type='function',
             function=ToolCallFunction(
                 name=tc['function']['name'],
-                arguments=tc['function']['arguments']
+                arguments=_normalize_tool_call_arguments(tc['function']['arguments'])
             )
         )
         for tc in tool_calls
     ]
 
+
+def _normalize_tool_call_arguments(arguments: Any) -> Any:
+    """Strip trailing streaming artifacts so arguments remain valid JSON strings."""
+    if not arguments or not isinstance(arguments, str):
+        return arguments
+
+    decoder = json.JSONDecoder()
+    try:
+        obj, end = decoder.raw_decode(arguments)
+    except json.JSONDecodeError:
+        return arguments
+
+    remainder = arguments[end:].strip()
+    if remainder:
+        try:
+            return json.dumps(obj)
+        except (TypeError, ValueError):
+            return arguments
+
+    return arguments
+
 async def _execute_tool_calls(
     tool_calls: List[ToolCall],
     agent: Agent[Ctx, Any],
@@ -863,7 +892,8 @@ async def _execute_tool_calls(
                 tool_name=tool_call.function.name,
                 args=_try_parse_json(tool_call.function.arguments),
                 trace_id=state.trace_id,
-                run_id=state.run_id
+                run_id=state.run_id,
+                call_id=tool_call.id
             ))))
 
         try:
@@ -889,7 +919,8 @@ async def _execute_tool_calls(
                         trace_id=state.trace_id,
                         run_id=state.run_id,
                         status='error',
-                        tool_result={'error': 'tool_not_found'}
+                        tool_result={'error': 'tool_not_found'},
+                        call_id=tool_call.id
                     ))))
 
                 return {
@@ -923,7 +954,8 @@ async def _execute_tool_calls(
                         trace_id=state.trace_id,
                         run_id=state.run_id,
                         status='error',
-                        tool_result={'error': 'validation_error', 'details': e.errors()}
+                        tool_result={'error': 'validation_error', 'details': e.errors()},
+                        call_id=tool_call.id
                     ))))
 
                 return {
@@ -1017,7 +1049,7 @@ async def _execute_tool_calls(
             else:
                 timeout = None
             if timeout is None:
-                timeout = config.default_tool_timeout if config.default_tool_timeout is not None else 30.0
+                timeout = config.default_tool_timeout if config.default_tool_timeout is not None else 300.0
 
             # Merge additional context if provided through approval
             additional_context = approval_status.additional_context if approval_status else None
@@ -1061,7 +1093,8 @@ async def _execute_tool_calls(
                         trace_id=state.trace_id,
                         run_id=state.run_id,
                         status='timeout',
-                        tool_result={'error': 'timeout_error'}
+                        tool_result={'error': 'timeout_error'},
+                        call_id=tool_call.id
                     ))))
 
                 return {
@@ -1113,7 +1146,8 @@ async def _execute_tool_calls(
                     trace_id=state.trace_id,
                     run_id=state.run_id,
                     tool_result=tool_result,
-                    status='success'
+                    status='success',
+                    call_id=tool_call.id
                 ))))
 
             # Check for handoff
@@ -1151,7 +1185,8 @@ async def _execute_tool_calls(
                     trace_id=state.trace_id,
                     run_id=state.run_id,
                     status='error',
-                    tool_result={'error': 'execution_error', 'detail': str(error)}
+                    tool_result={'error': 'execution_error', 'detail': str(error)},
+                    call_id=tool_call.id
                 ))))
 
             return {

jaf/core/parallel_agents.py

@@ -0,0 +1,339 @@
+"""
+Parallel Agent Execution for JAF Framework.
+
+This module provides functionality to execute multiple sub-agents in parallel groups,
+allowing for coordinated parallel execution with configurable grouping and result aggregation.
+"""
+
+import asyncio
+import json
+from dataclasses import dataclass
+from typing import Any, Dict, List, Optional, Union, Callable, TypeVar
+
+from .types import (
+    Agent,
+    Tool,
+    ToolSchema,
+    ToolSource,
+    RunConfig,
+    RunState,
+    RunResult,
+    Message,
+    ContentRole,
+    generate_run_id,
+    generate_trace_id,
+)
+from .agent_tool import create_agent_tool, AgentToolInput
+
+Ctx = TypeVar('Ctx')
+Out = TypeVar('Out')
+
+
+@dataclass
+class ParallelAgentGroup:
+    """Configuration for a group of agents to be executed in parallel."""
+    name: str
+    agents: List[Agent[Ctx, Out]]
+    shared_input: bool = True  # Whether all agents receive the same input
+    result_aggregation: str = "combine"  # "combine", "first", "majority", "custom"
+    custom_aggregator: Optional[Callable[[List[str]], str]] = None
+    timeout: Optional[float] = None
+    metadata: Optional[Dict[str, Any]] = None
+
+
+@dataclass
+class ParallelExecutionConfig:
+    """Configuration for parallel agent execution."""
+    groups: List[ParallelAgentGroup]
+    inter_group_execution: str = "sequential"  # "sequential" or "parallel"
+    global_timeout: Optional[float] = None
+    preserve_session: bool = False
+
+
+class ParallelAgentsTool:
+    """Tool that executes multiple agent groups in parallel."""
+    
+    def __init__(
+        self,
+        config: ParallelExecutionConfig,
+        tool_name: str = "execute_parallel_agents",
+        tool_description: str = "Execute multiple agents in parallel groups"
+    ):
+        self.config = config
+        self.tool_name = tool_name
+        self.tool_description = tool_description
+        
+        # Create tool schema
+        self.schema = ToolSchema(
+            name=tool_name,
+            description=tool_description,
+            parameters=AgentToolInput,
+            timeout=config.global_timeout
+        )
+        self.source = ToolSource.NATIVE
+        self.metadata = {"source": "parallel_agents", "groups": len(config.groups)}
+    
+    async def execute(self, args: AgentToolInput, context: Ctx) -> str:
+        """Execute all configured agent groups."""
+        try:
+            if self.config.inter_group_execution == "parallel":
+                # Execute all groups in parallel
+                group_results = await asyncio.gather(*[
+                    self._execute_group(group, args.input, context) 
+                    for group in self.config.groups
+                ])
+            else:
+                # Execute groups sequentially
+                group_results = []
+                for group in self.config.groups:
+                    result = await self._execute_group(group, args.input, context)
+                    group_results.append(result)
+            
+            # Combine results from all groups
+            final_result = {
+                "parallel_execution_results": {
+                    group.name: result for group, result in zip(self.config.groups, group_results)
+                },
+                "execution_mode": self.config.inter_group_execution,
+                "total_groups": len(self.config.groups)
+            }
+            
+            return json.dumps(final_result, indent=2)
+            
+        except Exception as e:
+            return json.dumps({
+                "error": "parallel_execution_failed",
+                "message": f"Failed to execute parallel agents: {str(e)}",
+                "groups_attempted": len(self.config.groups)
+            })
+    
+    async def _execute_group(
+        self, 
+        group: ParallelAgentGroup, 
+        input_text: str, 
+        context: Ctx
+    ) -> Dict[str, Any]:
+        """Execute a single group of agents in parallel."""
+        try:
+            # Create agent tools for all agents in the group
+            agent_tools = []
+            for agent in group.agents:
+                tool = create_agent_tool(
+                    agent=agent,
+                    tool_name=f"run_{agent.name.lower().replace(' ', '_')}",
+                    tool_description=f"Execute the {agent.name} agent",
+                    timeout=group.timeout,
+                    preserve_session=self.config.preserve_session
+                )
+                agent_tools.append((agent.name, tool))
+            
+            # Execute all agents in the group in parallel
+            if group.shared_input:
+                # All agents get the same input
+                tasks = [
+                    tool.execute(AgentToolInput(input=input_text), context)
+                    for _, tool in agent_tools
+                ]
+            else:
+                # This could be extended to support different inputs per agent
+                tasks = [
+                    tool.execute(AgentToolInput(input=input_text), context)
+                    for _, tool in agent_tools
+                ]
+            
+            # Execute with timeout if specified
+            if group.timeout:
+                results = await asyncio.wait_for(
+                    asyncio.gather(*tasks, return_exceptions=True),
+                    timeout=group.timeout
+                )
+            else:
+                results = await asyncio.gather(*tasks, return_exceptions=True)
+            
+            # Process results
+            agent_results = {}
+            for (agent_name, _), result in zip(agent_tools, results):
+                if isinstance(result, Exception):
+                    agent_results[agent_name] = {
+                        "error": True,
+                        "message": str(result),
+                        "type": type(result).__name__
+                    }
+                else:
+                    agent_results[agent_name] = {
+                        "success": True,
+                        "result": result
+                    }
+            
+            # Apply result aggregation
+            aggregated_result = self._aggregate_results(group, agent_results)
+            
+            return {
+                "group_name": group.name,
+                "agent_count": len(group.agents),
+                "individual_results": agent_results,
+                "aggregated_result": aggregated_result,
+                "execution_time_ms": None  # Could be added with timing
+            }
+            
+        except asyncio.TimeoutError:
+            return {
+                "group_name": group.name,
+                "error": "timeout",
+                "message": f"Group {group.name} execution timed out after {group.timeout} seconds",
+                "agent_count": len(group.agents)
+            }
+        except Exception as e:
+            return {
+                "group_name": group.name,
+                "error": "execution_failed",
+                "message": str(e),
+                "agent_count": len(group.agents)
+            }
+    
+    def _aggregate_results(
+        self, 
+        group: ParallelAgentGroup, 
+        agent_results: Dict[str, Any]
+    ) -> Union[str, Dict[str, Any]]:
+        """Aggregate results from parallel agent execution."""
+        successful_results = [
+            result["result"] for result in agent_results.values() 
+            if result.get("success") and "result" in result
+        ]
+        
+        if not successful_results:
+            return {"error": "no_successful_results", "message": "All agents failed"}
+        
+        if group.result_aggregation == "first":
+            return successful_results[0]
+        elif group.result_aggregation == "combine":
+            return {
+                "combined_results": successful_results,
+                "result_count": len(successful_results)
+            }
+        elif group.result_aggregation == "majority":
+            # Simple majority logic - could be enhanced
+            if len(successful_results) >= len(group.agents) // 2 + 1:
+                return successful_results[0]  # Return first as majority representative
+            else:
+                return {"error": "no_majority", "results": successful_results}
+        elif group.result_aggregation == "custom" and group.custom_aggregator:
+            try:
+                return group.custom_aggregator(successful_results)
+            except Exception as e:
+                return {"error": "custom_aggregation_failed", "message": str(e)}
+        else:
+            return {"combined_results": successful_results}
+
+
+def create_parallel_agents_tool(
+    groups: List[ParallelAgentGroup],
+    tool_name: str = "execute_parallel_agents",
+    tool_description: str = "Execute multiple agents in parallel groups",
+    inter_group_execution: str = "sequential",
+    global_timeout: Optional[float] = None,
+    preserve_session: bool = False
+) -> Tool:
+    """
+    Create a tool that executes multiple agent groups in parallel.
+    
+    Args:
+        groups: List of parallel agent groups to execute
+        tool_name: Name of the tool
+        tool_description: Description of the tool
+        inter_group_execution: How to execute groups ("sequential" or "parallel")
+        global_timeout: Global timeout for all executions
+        preserve_session: Whether to preserve session across agent calls
+        
+    Returns:
+        A Tool that can execute parallel agent groups
+    """
+    config = ParallelExecutionConfig(
+        groups=groups,
+        inter_group_execution=inter_group_execution,
+        global_timeout=global_timeout,
+        preserve_session=preserve_session
+    )
+    
+    return ParallelAgentsTool(config, tool_name, tool_description)
+
+
+def create_simple_parallel_tool(
+    agents: List[Agent],
+    group_name: str = "parallel_group",
+    tool_name: str = "execute_parallel_agents",
+    shared_input: bool = True,
+    result_aggregation: str = "combine",
+    timeout: Optional[float] = None
+) -> Tool:
+    """
+    Create a simple parallel agents tool from a list of agents.
+    
+    Args:
+        agents: List of agents to execute in parallel
+        group_name: Name for the parallel group
+        tool_name: Name of the tool
+        shared_input: Whether all agents receive the same input
+        result_aggregation: How to aggregate results ("combine", "first", "majority")
+        timeout: Timeout for parallel execution
+        
+    Returns:
+        A Tool that executes all agents in parallel
+    """
+    group = ParallelAgentGroup(
+        name=group_name,
+        agents=agents,
+        shared_input=shared_input,
+        result_aggregation=result_aggregation,
+        timeout=timeout
+    )
+    
+    return create_parallel_agents_tool([group], tool_name=tool_name)
+
+
+# Convenience functions for common parallel execution patterns
+
+def create_language_specialists_tool(
+    language_agents: Dict[str, Agent],
+    tool_name: str = "consult_language_specialists",
+    timeout: Optional[float] = 300.0
+) -> Tool:
+    """Create a tool that consults multiple language specialists in parallel."""
+    group = ParallelAgentGroup(
+        name="language_specialists",
+        agents=list(language_agents.values()),
+        shared_input=True,
+        result_aggregation="combine",
+        timeout=timeout,
+        metadata={"languages": list(language_agents.keys())}
+    )
+    
+    return create_parallel_agents_tool(
+        [group], 
+        tool_name=tool_name,
+        tool_description="Consult multiple language specialists in parallel"
+    )
+
+
+def create_domain_experts_tool(
+    expert_agents: Dict[str, Agent],
+    tool_name: str = "consult_domain_experts",
+    result_aggregation: str = "combine",
+    timeout: Optional[float] = 60.0
+) -> Tool:
+    """Create a tool that consults multiple domain experts in parallel."""
+    group = ParallelAgentGroup(
+        name="domain_experts",
+        agents=list(expert_agents.values()),
+        shared_input=True,
+        result_aggregation=result_aggregation,
+        timeout=timeout,
+        metadata={"domains": list(expert_agents.keys())}
+    )
+    
+    return create_parallel_agents_tool(
+        [group],
+        tool_name=tool_name,
+        tool_description="Consult multiple domain experts in parallel"
+    )

jaf/core/streaming.py

@@ -209,20 +209,37 @@ async def run_streaming(
         trace_id=initial_state.trace_id
     )
 
-    tool_call_ids = {} # To map tool calls to their IDs
+    tool_call_ids: Dict[str, str] = {}  # Map call_id -> tool_name for in-flight tool calls
 
     def event_handler(event: TraceEvent) -> None:
         """Handle trace events and put them into the queue."""
         nonlocal tool_call_ids
         streaming_event = None
+        payload = event.data
+
+        def _get_event_value(keys: List[str]) -> Any:
+            for key in keys:
+                if isinstance(payload, dict) and key in payload:
+                    return payload[key]
+                if hasattr(payload, key):
+                    return getattr(payload, key)
+            return None
+
         if event.type == 'tool_call_start':
-            # Generate a unique ID for the tool call
-            call_id = f"call_{uuid.uuid4().hex[:8]}"
-            tool_call_ids[event.data.tool_name] = call_id
-            
+            tool_name = _get_event_value(['tool_name', 'toolName']) or 'unknown'
+            args = _get_event_value(['args', 'arguments'])
+            call_id = _get_event_value(['call_id', 'tool_call_id', 'toolCallId'])
+
+            if not call_id:
+                call_id = f"call_{uuid.uuid4().hex[:8]}"
+                if isinstance(payload, dict):
+                    payload['call_id'] = call_id
+
+            tool_call_ids[call_id] = tool_name
+
             tool_call = StreamingToolCall(
-                tool_name=event.data.tool_name,
-                arguments=event.data.args,
+                tool_name=tool_name,
+                arguments=args,
                 call_id=call_id,
                 status='started'
             )
@@ -233,18 +250,26 @@ async def run_streaming(
                 trace_id=initial_state.trace_id
             )
         elif event.type == 'tool_call_end':
-            if event.data.tool_name not in tool_call_ids:
-                raise RuntimeError(
-                    f"Tool call end event received for unknown tool '{event.data.tool_name}'. "
-                    f"Known tool calls: {list(tool_call_ids.keys())}. "
-                    f"This may indicate a missing tool_call_start event or a bug in the streaming implementation."
-                )
-            call_id = tool_call_ids[event.data.tool_name]
+            tool_name = _get_event_value(['tool_name', 'toolName']) or 'unknown'
+            call_id = _get_event_value(['call_id', 'tool_call_id', 'toolCallId'])
+
+            if not call_id:
+                # Fallback to locate a pending tool call with the same tool name
+                matching_call_id = next((cid for cid, name in tool_call_ids.items() if name == tool_name), None)
+                if matching_call_id:
+                    call_id = matching_call_id
+                else:
+                    raise RuntimeError(
+                        f"Tool call end event received for unknown tool '{tool_name}'. "
+                        f"Pending call IDs: {list(tool_call_ids.keys())}."
+                    )
+
+            tool_call_ids.pop(call_id, None)
             tool_result = StreamingToolResult(
-                tool_name=event.data.tool_name,
+                tool_name=tool_name,
                 call_id=call_id,
-                result=event.data.result,
-                status=event.data.status or 'completed'
+                result=_get_event_value(['result']),
+                status=_get_event_value(['status']) or 'completed'
             )
             streaming_event = StreamingEvent(
                 type=StreamingEventType.TOOL_RESULT,

jaf/core/tracing.py

@@ -10,6 +10,7 @@ import json
 import time
 from datetime import datetime
 from typing import Any, Dict, List, Optional, Protocol
+import uuid
 
 from opentelemetry import trace
 from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
@@ -419,11 +420,24 @@ class LangfuseTraceCollector:
                     for i in range(len(messages) - 1, -1, -1):
                         msg = messages[i]
                         
+                        # Extract message data comprehensively
+                        msg_data = {}
+                        
                         if isinstance(msg, dict):
                             role = msg.get("role")
                             content = msg.get("content", "")
+                            # Capture all additional fields from dict messages
+                            msg_data = {
+                                "role": role,
+                                "content": content,
+                                "tool_calls": msg.get("tool_calls"),
+                                "tool_call_id": msg.get("tool_call_id"),
+                                "name": msg.get("name"),
+                                "function_call": msg.get("function_call"),
+                                "timestamp": msg.get("timestamp", datetime.now().isoformat())
+                            }
                         elif hasattr(msg, 'role'):
-                            role = msg.role
+                            role = getattr(msg, 'role', None)
                             content = getattr(msg, 'content', "")
                             # Handle both string content and complex content structures
                             if not isinstance(content, str):
@@ -436,9 +450,25 @@ class LangfuseTraceCollector:
                                         content = str(content)
                                 else:
                                     content = str(content)
+                            
+                            # Capture all additional fields from object messages
+                            msg_data = {
+                                "role": role,
+                                "content": content,
+                                "tool_calls": getattr(msg, 'tool_calls', None),
+                                "tool_call_id": getattr(msg, 'tool_call_id', None),
+                                "name": getattr(msg, 'name', None),
+                                "function_call": getattr(msg, 'function_call', None),
+                                "timestamp": getattr(msg, 'timestamp', datetime.now().isoformat())
+                            }
                         else:
+                            # Handle messages that don't have expected structure
+                            print(f"[LANGFUSE DEBUG] Skipping message with unexpected structure: {type(msg)}")
                             continue
                         
+                        # Clean up None values from msg_data
+                        msg_data = {k: v for k, v in msg_data.items() if v is not None}
+                        
                         # If we haven't found the current user message yet and this is a user message
                         if not current_user_message_found and (role == "user" or role == 'user'):
                             user_query = content
@@ -446,15 +476,29 @@ class LangfuseTraceCollector:
                             print(f"[LANGFUSE DEBUG] Found current user query: {user_query}")
                         elif current_user_message_found:
                             # Add to conversation history (excluding the current user message)
-                            conversation_history.insert(0, {
-                                "role": role,
-                                "content": content,
-                                "timestamp": datetime.now().isoformat() if not hasattr(msg, 'timestamp') else getattr(msg, 'timestamp', datetime.now().isoformat())
-                            })
+                            # Include ALL message types: assistant, tool, system, function, etc.
+                            conversation_history.insert(0, msg_data)
+                            print(f"[LANGFUSE DEBUG] Added to conversation history: role={role}, content_length={len(str(content))}, has_tool_calls={bool(msg_data.get('tool_calls'))}")
                 
                 print(f"[LANGFUSE DEBUG] Final extracted - user_query: {user_query}, user_id: {user_id}")
                 print(f"[LANGFUSE DEBUG] Conversation history length: {len(conversation_history)}")
                 
+                # Debug: Log the roles and types captured in conversation history
+                if conversation_history:
+                    roles_summary = {}
+                    for msg in conversation_history:
+                        role = msg.get("role", "unknown")
+                        roles_summary[role] = roles_summary.get(role, 0) + 1
+                    print(f"[LANGFUSE DEBUG] Conversation history roles breakdown: {roles_summary}")
+                    
+                    # Log first few messages for verification
+                    for i, msg in enumerate(conversation_history[:3]):
+                        role = msg.get("role", "unknown")
+                        content_preview = str(msg.get("content", ""))[:100]
+                        has_tool_calls = bool(msg.get("tool_calls"))
+                        has_tool_call_id = bool(msg.get("tool_call_id"))
+                        print(f"[LANGFUSE DEBUG] History msg {i}: role={role}, content='{content_preview}...', tool_calls={has_tool_calls}, tool_call_id={has_tool_call_id}")
+                
                 # Create comprehensive input data for the trace
                 trace_input = {
                     "user_query": user_query,
@@ -495,6 +539,7 @@ class LangfuseTraceCollector:
                     print(f"[LANGFUSE] Ending trace for run: {trace_id}")
                     
                     # Update the trace metadata with final tool calls and results
+                    conversation_history = getattr(self.trace_spans[trace_id], '_conversation_history', [])
                     final_metadata = {
                         "framework": "jaf",
                         "event_type": "run_end",
@@ -502,7 +547,7 @@ class LangfuseTraceCollector:
                         "user_query": getattr(self.trace_spans[trace_id], '_user_query', None),
                         "user_id": getattr(self.trace_spans[trace_id], '_user_id', None),
                         "agent_name": event.data.get("agent_name", "analytics_agent_jaf"),
-                        "conversation_history": getattr(self.trace_spans[trace_id], '_conversation_history', []),
+                        "conversation_history": conversation_history,
                         "tool_calls": self.trace_tool_calls.get(trace_id, []),
                         "tool_results": self.trace_tool_results.get(trace_id, [])
                     }
@@ -608,28 +653,36 @@ class LangfuseTraceCollector:
                 # Start a span for tool calls with detailed input information
                 tool_name = event.data.get('tool_name', 'unknown')
                 tool_args = event.data.get("args", {})
+                call_id = event.data.get("call_id")
+                if not call_id:
+                    call_id = f"{tool_name}-{uuid.uuid4().hex[:8]}"
+                    try:
+                        event.data["call_id"] = call_id
+                    except TypeError:
+                        # event.data may be immutable; log and rely on synthetic ID tracking downstream
+                        print(f"[LANGFUSE] Generated synthetic call_id for tool start: {call_id}")
                 
-                print(f"[LANGFUSE] Starting span for tool call: {tool_name}")
+                print(f"[LANGFUSE] Starting span for tool call: {tool_name} ({call_id})")
                 
                 # Track this tool call for the trace
                 tool_call_data = {
                     "tool_name": tool_name,
                     "arguments": tool_args,
-                    "call_id": event.data.get("call_id"),
+                    "call_id": call_id,
                     "timestamp": datetime.now().isoformat()
                 }
                 
                 # Ensure trace_id exists in tracking
                 if trace_id not in self.trace_tool_calls:
                     self.trace_tool_calls[trace_id] = []
-                
+
                 self.trace_tool_calls[trace_id].append(tool_call_data)
                 
                 # Create comprehensive input data for the tool call
                 tool_input = {
                     "tool_name": tool_name,
                     "arguments": tool_args,
-                    "call_id": event.data.get("call_id"),
+                    "call_id": call_id,
                     "timestamp": datetime.now().isoformat()
                 }
                 
@@ -638,7 +691,7 @@ class LangfuseTraceCollector:
                     input=tool_input,
                     metadata={
                         "tool_name": tool_name,
-                        "call_id": event.data.get("call_id"),
+                        "call_id": call_id,
                         "framework": "jaf",
                         "event_type": "tool_call"
                     }
@@ -652,14 +705,15 @@ class LangfuseTraceCollector:
                 if span_id in self.active_spans:
                     tool_name = event.data.get('tool_name', 'unknown')
                     tool_result = event.data.get("result")
+                    call_id = event.data.get("call_id")
                     
-                    print(f"[LANGFUSE] Ending span for tool call: {tool_name}")
+                    print(f"[LANGFUSE] Ending span for tool call: {tool_name} ({call_id})")
                     
                     # Track this tool result for the trace
                     tool_result_data = {
                         "tool_name": tool_name,
                         "result": tool_result,
-                        "call_id": event.data.get("call_id"),
+                        "call_id": call_id,
                         "timestamp": datetime.now().isoformat(),
                         "status": event.data.get("status", "completed"),
                         "tool_result": event.data.get("tool_result")
@@ -674,7 +728,7 @@ class LangfuseTraceCollector:
                     tool_output = {
                         "tool_name": tool_name,
                         "result": tool_result,
-                        "call_id": event.data.get("call_id"),
+                        "call_id": call_id,
                         "timestamp": datetime.now().isoformat(),
                         "status": event.data.get("status", "completed")
                     }
@@ -685,7 +739,7 @@ class LangfuseTraceCollector:
                         output=tool_output,
                         metadata={
                             "tool_name": tool_name,
-                            "call_id": event.data.get("call_id"),
+                            "call_id": call_id,
                             "result_length": len(str(tool_result)) if tool_result else 0,
                             "framework": "jaf",
                             "event_type": "tool_call_end"
@@ -747,6 +801,9 @@ class LangfuseTraceCollector:
         
         # Use consistent identifiers that don't depend on timestamp
         if event.type.startswith('tool_call'):
+            call_id = event.data.get('call_id') or event.data.get('tool_call_id')
+            if call_id:
+                return f"tool-{trace_id}-{call_id}"
             tool_name = event.data.get('tool_name') or event.data.get('toolName', 'unknown')
             return f"tool-{tool_name}-{trace_id}"
         elif event.type.startswith('llm_call'):

jaf/core/types.py

@@ -490,6 +490,7 @@ class RunStartEventData:
     session_id: Optional[str] = None
     context: Optional[Any] = None
     messages: Optional[List[Message]] = None
+    agent_name: Optional[str] = None
 
 @dataclass(frozen=True)
 class RunStartEvent:
@@ -540,11 +541,12 @@ class ToolCallStartEventData:
     args: Any
     trace_id: TraceId
     run_id: RunId
+    call_id: Optional[str] = None
 
 @dataclass(frozen=True)
 class ToolCallStartEvent:
     type: Literal['tool_call_start'] = 'tool_call_start'
-    data: ToolCallStartEventData = field(default_factory=lambda: ToolCallStartEventData("", None, TraceId(""), RunId("")))
+    data: ToolCallStartEventData = field(default_factory=lambda: ToolCallStartEventData("", None, TraceId(""), RunId(""), None))
 
 @dataclass(frozen=True)
 class ToolCallEndEventData:
@@ -555,11 +557,12 @@ class ToolCallEndEventData:
     run_id: RunId
     tool_result: Optional[Any] = None
     status: Optional[str] = None
+    call_id: Optional[str] = None
 
 @dataclass(frozen=True)
 class ToolCallEndEvent:
     type: Literal['tool_call_end'] = 'tool_call_end'
-    data: ToolCallEndEventData = field(default_factory=lambda: ToolCallEndEventData("", "", TraceId(""), RunId("")))
+    data: ToolCallEndEventData = field(default_factory=lambda: ToolCallEndEventData("", "", TraceId(""), RunId(""), None, None))
 
 @dataclass(frozen=True)
 class HandoffEventData:
@@ -709,5 +712,5 @@ class RunConfig(Generic[Ctx]):
     on_event: Optional[Callable[[TraceEvent], None]] = None
     memory: Optional['MemoryConfig'] = None
     conversation_id: Optional[str] = None
-    default_tool_timeout: Optional[float] = 30.0  # Default timeout for tool execution in seconds
-    approval_storage: Optional['ApprovalStorage'] = None  # Storage for approval decisions
+    default_tool_timeout: Optional[float] = 300.0  # Default timeout for tool execution in seconds
+    approval_storage: Optional['ApprovalStorage'] = None  # Storage for approval decisions

jaf/server/server.py

@@ -533,12 +533,12 @@ def create_jaf_server(config: ServerConfig[Ctx]) -> FastAPI:
             async def event_stream():
                 try:
                     # Send initial metadata
-                    yield f"event: stream_start\ndata: {json.dumps({
+                    yield f"""event: stream_start data: {json.dumps({
                         'runId': str(initial_state.run_id),
                         'traceId': str(initial_state.trace_id),
                         'conversationId': conversation_id,
                         'agent': request.agent_name
-                    })}\n\n"
+                    })}"""
                     
                     # Stream events from the engine
                     async for event in run_streaming(initial_state, run_config_with_memory):

{jaf_py-2.4.3.dist-info → jaf_py-2.4.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: jaf-py
-Version: 2.4.3
+Version: 2.4.5
 Summary: A purely functional agent framework with immutable state and composable tools - Python implementation
 Author: JAF Contributors
 Maintainer: JAF Contributors

{jaf_py-2.4.3.dist-info → jaf_py-2.4.5.dist-info}/RECORD

@@ -38,21 +38,22 @@ jaf/a2a/tests/test_client.py,sha256=L5h7DtQRVlULiRhRLtrmaCoYdvmbXsgLTy3QQ6KgmNM,
 jaf/a2a/tests/test_integration.py,sha256=I7LdgwN99mAOljM9kYtK7dGMMntTSWKMw_oLOcJjinU,18454
 jaf/a2a/tests/test_protocol.py,sha256=He3vGlBfIazpppAnuSybutrvjIN3VGxEleAohrVd9hc,23287
 jaf/a2a/tests/test_types.py,sha256=PgRjDVJrHSXuu05z0B5lsSUUY5qEdQLFJbLBIExyVgI,18384
-jaf/core/__init__.py,sha256=rBvP_7TGbJICDJnA7a3qyX8yQErCDWaGAn5WzpyH4gU,1339
+jaf/core/__init__.py,sha256=PIGKm8n6OQ8jcXRS0Hn3_Zsl8m2qX91N80YJoLCJ4eU,1762
 jaf/core/agent_tool.py,sha256=tfLNaTIcOZ0dR9GBP1AHLPkLExm_dLbURnVIN4R84FQ,11806
 jaf/core/analytics.py,sha256=zFHIWqWal0bbEFCmJDc4DKeM0Ja7b_D19PqVaBI12pA,23338
 jaf/core/composition.py,sha256=IVxRO1Q9nK7JRH32qQ4p8WMIUu66BhqPNrlTNMGFVwE,26317
-jaf/core/engine.py,sha256=uijM1N-W-OG1oelMtV4m-o2cdrBziLU75jn8JOZBaWA,50631
+jaf/core/engine.py,sha256=bjHNn8MoE3o0BAuBCMBY3EjtpMckjWlBotJ-oinfSZ0,52111
 jaf/core/errors.py,sha256=5fwTNhkojKRQ4wZj3lZlgDnAsrYyjYOwXJkIr5EGNUc,5539
+jaf/core/parallel_agents.py,sha256=ahwYoTnkrF4xQgV-hjc5sUaWhQWQFENMZG5riNa_Ieg,12165
 jaf/core/performance.py,sha256=jedQmTEkrKMD6_Aw1h8PdG-5TsdYSFFT7Or6k5dmN2g,9974
 jaf/core/proxy.py,sha256=_WM3cpRlSQLYpgSBrnY30UPMe2iZtlqDQ65kppE-WY0,4609
 jaf/core/proxy_helpers.py,sha256=i7a5fAX9rLmO4FMBX51-yRkTFwfWedzQNgnLmeLUd_A,4370
 jaf/core/state.py,sha256=NMtYTpUYa64m1Kte6lD8LGnF2bl69HAcdgXH6f-M97c,5650
-jaf/core/streaming.py,sha256=c5o9iqpjoYV2LrUpG6qLWCYrWcP-DCcZsvMbyqKunp8,16089
+jaf/core/streaming.py,sha256=h_lYHQA9ee_D5QsDO9-Vhevgi7rFXPslPzd9605AJGo,17034
 jaf/core/tool_results.py,sha256=-bTOqOX02lMyslp5Z4Dmuhx0cLd5o7kgR88qK2HO_sw,11323
 jaf/core/tools.py,sha256=84N9A7QQ3xxcOs2eUUot3nmCnt5i7iZT9VwkuzuFBxQ,16274
-jaf/core/tracing.py,sha256=pOswPTvhIYXj88kxMSJDySfbcnROU3eB4dooVl1e_18,36604
-jaf/core/types.py,sha256=_e7F368BwzdZ2BZuakSp5Bw7gzi8xYD0LlOIy2S25Ds,23061
+jaf/core/tracing.py,sha256=iuVgykFUSkoBjem1k6jdVLrhRZzJn-avyxc_6W9BXPI,40159
+jaf/core/types.py,sha256=8TA5cyNAc_rp5Tn-zmvt6rMi_0atRXbewpaiB5Ss7-g,23186
 jaf/core/workflows.py,sha256=Ul-82gzjIXtkhnSMSPv-8igikjkMtW1EBo9yrfodtvI,26294
 jaf/memory/__init__.py,sha256=-L98xlvihurGAzF0DnXtkueDVvO_wV2XxxEwAWdAj50,1400
 jaf/memory/approval_storage.py,sha256=HHZ_b57kIthdR53QE5XNSII9xy1Cg-1cFUCSAZ8A4Rk,11083
@@ -73,7 +74,7 @@ jaf/providers/mcp.py,sha256=WxcC8gUFpDBBYyhorMcc1jHq3xMDMBtnwyRPthfL0S0,13074
 jaf/providers/model.py,sha256=NJTa-1k0EruDdLf2HS1ZdDpFJhHXzzfQyXAbJx9kZVc,25468
 jaf/server/__init__.py,sha256=fMPnLZBRm6t3yQrr7-PnoHAQ8qj9o6Z1AJLM1M6bIS0,392
 jaf/server/main.py,sha256=CTb0ywbPIq9ELfay5MKChVR7BpIQOoEbPjPfpzo2aBQ,2152
-jaf/server/server.py,sha256=LooAxxdeitwZHJff-UTE3L7zM6pSNqixMipOgspL5SA,38787
+jaf/server/server.py,sha256=K8XKNyadP_YqkCRSK9jCVZh52d2_IbHp_jHkKzBeB9Q,38786
 jaf/server/types.py,sha256=Gg8z1bkA7IYg94lp31iB92-4VkJr9syKA41uVCfNZBc,10544
 jaf/utils/__init__.py,sha256=4Lte7HPIpmEuGvWd9lSH9gljV11wy-yNFjECPhcejAY,1236
 jaf/utils/attachments.py,sha256=9xNzzQanCwJnBR1L6P79YQtbuRupiDluDn46SYUlHok,13542
@@ -84,9 +85,9 @@ jaf/visualization/functional_core.py,sha256=zedMDZbvjuOugWwnh6SJ2stvRNQX1Hlkb9Ab
 jaf/visualization/graphviz.py,sha256=WTOM6UP72-lVKwI4_SAr5-GCC3ouckxHv88ypCDQWJ0,12056
 jaf/visualization/imperative_shell.py,sha256=GpMrAlMnLo2IQgyB2nardCz09vMvAzaYI46MyrvJ0i4,2593
 jaf/visualization/types.py,sha256=QQcbVeQJLuAOXk8ynd08DXIS-PVCnv3R-XVE9iAcglw,1389
-jaf_py-2.4.3.dist-info/licenses/LICENSE,sha256=LXUQBJxdyr-7C4bk9cQBwvsF_xwA-UVstDTKabpcjlI,1063
-jaf_py-2.4.3.dist-info/METADATA,sha256=b3kDMPBoPR_Z44jyVnZHnpHM-bSDTkuXeSegWXsUSHw,27712
-jaf_py-2.4.3.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-jaf_py-2.4.3.dist-info/entry_points.txt,sha256=OtIJeNJpb24kgGrqRx9szGgDx1vL9ayq8uHErmu7U5w,41
-jaf_py-2.4.3.dist-info/top_level.txt,sha256=Xu1RZbGaM4_yQX7bpalo881hg7N_dybaOW282F15ruE,4
-jaf_py-2.4.3.dist-info/RECORD,,
+jaf_py-2.4.5.dist-info/licenses/LICENSE,sha256=LXUQBJxdyr-7C4bk9cQBwvsF_xwA-UVstDTKabpcjlI,1063
+jaf_py-2.4.5.dist-info/METADATA,sha256=9oUGQOOBTFoMdCVPfnyC9ucxrBXWzvporO5hDuDXkjA,27712
+jaf_py-2.4.5.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+jaf_py-2.4.5.dist-info/entry_points.txt,sha256=OtIJeNJpb24kgGrqRx9szGgDx1vL9ayq8uHErmu7U5w,41
+jaf_py-2.4.5.dist-info/top_level.txt,sha256=Xu1RZbGaM4_yQX7bpalo881hg7N_dybaOW282F15ruE,4
+jaf_py-2.4.5.dist-info/RECORD,,