kubiya-control-plane-api 0.1.0__py3-none-any.whl → 0.3.4__py3-none-any.whl

control_plane_api/app/workflows/team_execution.py

@@ -0,0 +1,399 @@
+"""Team execution workflow for Temporal"""
+
+from dataclasses import dataclass, field
+from datetime import timedelta
+from typing import Optional, List, Dict, Any
+from temporalio import workflow
+import asyncio
+
+with workflow.unsafe.imports_passed_through():
+    from control_plane_api.app.activities.team_activities import (
+        get_team_agents,
+        execute_team_coordination,
+        update_execution_status,
+        ActivityGetTeamAgentsInput,
+        ActivityExecuteTeamInput,
+        ActivityUpdateExecutionInput,
+    )
+
+
+@dataclass
+class TeamExecutionInput:
+    """Input for team execution workflow"""
+    execution_id: str
+    team_id: str
+    organization_id: str
+    prompt: str
+    system_prompt: Optional[str] = None
+    model_id: Optional[str] = None
+    model_config: dict = None
+    team_config: dict = None
+    mcp_servers: dict = None
+    user_metadata: dict = None
+
+    def __post_init__(self):
+        if self.model_config is None:
+            self.model_config = {}
+        if self.team_config is None:
+            self.team_config = {}
+        if self.mcp_servers is None:
+            self.mcp_servers = {}
+        if self.user_metadata is None:
+            self.user_metadata = {}
+
+
+@dataclass
+class ChatMessage:
+    """Represents a message in the conversation"""
+    role: str  # "user", "assistant", "system", "tool"
+    content: str
+    timestamp: str
+    tool_name: Optional[str] = None
+    tool_input: Optional[Dict[str, Any]] = None
+    tool_output: Optional[Dict[str, Any]] = None
+    # User attribution for messages
+    user_id: Optional[str] = None
+    user_name: Optional[str] = None
+    user_email: Optional[str] = None
+    user_avatar: Optional[str] = None
+
+
+@dataclass
+class ExecutionState:
+    """Current state of the execution for queries"""
+    status: str  # "pending", "running", "waiting_for_input", "completed", "failed"
+    messages: List[ChatMessage] = field(default_factory=list)
+    current_response: str = ""
+    error_message: Optional[str] = None
+    usage: Dict[str, Any] = field(default_factory=dict)
+    metadata: Dict[str, Any] = field(default_factory=dict)
+    is_waiting_for_input: bool = False
+    should_complete: bool = False
+
+
+@workflow.defn
+class TeamExecutionWorkflow:
+    """
+    Workflow for executing a team of agents with HITL support.
+
+    This workflow:
+    1. Gets team agents
+    2. Coordinates execution across agents
+    3. Aggregates results
+    4. Updates execution status
+    5. Supports queries for real-time state access
+    6. Supports signals for adding followup messages
+    """
+
+    def __init__(self) -> None:
+        """Initialize workflow state"""
+        self._state = ExecutionState(status="pending")
+        self._lock = asyncio.Lock()
+        self._new_message_count = 0
+        self._processed_message_count = 0
+
+    @workflow.query
+    def get_state(self) -> ExecutionState:
+        """Query handler: Get current execution state including messages and status"""
+        return self._state
+
+    @workflow.signal
+    async def add_message(self, message: ChatMessage) -> None:
+        """
+        Signal handler: Add a message to the conversation.
+        This allows clients to send followup messages while the workflow is running.
+        """
+        async with self._lock:
+            self._state.messages.append(message)
+            self._new_message_count += 1
+            self._state.is_waiting_for_input = False
+            workflow.logger.info(
+                f"Message added to team conversation",
+                extra={
+                    "role": message.role,
+                    "content_preview": message.content[:100] if message.content else "",
+                    "total_messages": len(self._state.messages)
+                }
+            )
+
+    @workflow.signal
+    async def mark_as_done(self) -> None:
+        """
+        Signal handler: Mark the workflow as complete.
+        """
+        async with self._lock:
+            self._state.should_complete = True
+            self._state.is_waiting_for_input = False
+            workflow.logger.info("Team workflow marked as done by user")
+
+    @workflow.run
+    async def run(self, input: TeamExecutionInput) -> dict:
+        """
+        Run the team execution workflow with HITL pattern.
+
+        This workflow implements a continuous conversation loop:
+        1. Process the initial user message
+        2. Execute team coordination and return response
+        3. Wait for user input (signals)
+        4. Process followup messages in a loop
+        5. Only complete when user explicitly marks as done
+
+        Args:
+            input: Workflow input with team execution details
+
+        Returns:
+            Team execution result dict
+        """
+        workflow.logger.info(
+            f"Starting team execution workflow with HITL pattern",
+            extra={
+                "execution_id": input.execution_id,
+                "team_id": input.team_id,
+                "organization_id": input.organization_id,
+            }
+        )
+
+        # Initialize state with user's initial message
+        self._state.messages.append(ChatMessage(
+            role="user",
+            content=input.prompt,
+            timestamp=workflow.now().isoformat(),
+        ))
+        self._state.status = "running"
+        self._new_message_count = 1
+        self._processed_message_count = 0
+
+        try:
+            # Step 1: Update execution status to running
+            await workflow.execute_activity(
+                update_execution_status,
+                ActivityUpdateExecutionInput(
+                    execution_id=input.execution_id,
+                    status="running",
+                    started_at=workflow.now().isoformat(),
+                    execution_metadata={
+                        "workflow_started": True,
+                        "hitl_enabled": True,
+                    },
+                ),
+                start_to_close_timeout=timedelta(seconds=30),
+            )
+
+            # Step 2: Get team agents once at the beginning
+            workflow.logger.info(
+                f"[WORKFLOW] About to call get_team_agents",
+                extra={
+                    "team_id": input.team_id,
+                    "organization_id": input.organization_id,
+                }
+            )
+
+            team_agents = await workflow.execute_activity(
+                get_team_agents,
+                ActivityGetTeamAgentsInput(
+                    team_id=input.team_id,
+                    organization_id=input.organization_id,
+                ),
+                start_to_close_timeout=timedelta(seconds=30),
+            )
+
+            workflow.logger.info(
+                f"[WORKFLOW] get_team_agents returned",
+                extra={
+                    "result": team_agents,
+                    "agents_count": len(team_agents.get("agents", [])) if team_agents else 0,
+                }
+            )
+
+            if not team_agents.get("agents"):
+                workflow.logger.error(
+                    f"[WORKFLOW] NO AGENTS RETURNED!",
+                    extra={
+                        "team_agents": team_agents,
+                        "team_id": input.team_id,
+                        "organization_id": input.organization_id,
+                    }
+                )
+                raise ValueError("No agents found in team")
+
+            # HITL Conversation Loop - Continue until user marks as done
+            conversation_turn = 0
+            while not self._state.should_complete:
+                conversation_turn += 1
+                workflow.logger.info(
+                    f"Starting team conversation turn {conversation_turn}",
+                    extra={"turn": conversation_turn, "message_count": len(self._state.messages)}
+                )
+
+                # Get the latest user message
+                latest_message = self._state.messages[-1] if self._state.messages else None
+                latest_prompt = latest_message.content if latest_message and latest_message.role == "user" else input.prompt
+
+                # Step 3: Execute team coordination
+                team_result = await workflow.execute_activity(
+                    execute_team_coordination,
+                    ActivityExecuteTeamInput(
+                        execution_id=input.execution_id,
+                        team_id=input.team_id,
+                        organization_id=input.organization_id,
+                        prompt=latest_prompt,
+                        system_prompt=input.system_prompt,
+                        agents=team_agents["agents"],
+                        team_config=input.team_config,
+                        mcp_servers=input.mcp_servers,  # Pass MCP servers
+                        session_id=input.execution_id,  # Use execution_id as session_id for 1:1 mapping
+                        user_id=input.user_metadata.get("user_id") if input.user_metadata else None,  # Extract user_id from JWT
+                    ),
+                    start_to_close_timeout=timedelta(minutes=30),  # Teams can take longer
+                )
+
+                # Update state with team response
+                if team_result.get("response"):
+                    async with self._lock:
+                        self._state.messages.append(ChatMessage(
+                            role="assistant",
+                            content=team_result["response"],
+                            timestamp=workflow.now().isoformat(),
+                        ))
+                        self._state.current_response = team_result["response"]
+                        self._processed_message_count += 1
+
+                # Update usage and metadata (accumulate across turns)
+                if team_result.get("usage"):
+                    current_usage = self._state.usage
+                    new_usage = team_result.get("usage", {})
+                    self._state.usage = {
+                        "input_tokens": current_usage.get("input_tokens", 0) + new_usage.get("input_tokens", 0),
+                        "output_tokens": current_usage.get("output_tokens", 0) + new_usage.get("output_tokens", 0),
+                        "total_tokens": current_usage.get("total_tokens", 0) + new_usage.get("total_tokens", 0),
+                    }
+
+                # Update metadata
+                self._state.metadata.update({
+                    "agent_count": len(team_agents["agents"]),
+                    "coordination_type": team_result.get("coordination_type"),
+                    "conversation_turns": conversation_turn,
+                })
+
+                # Check if team execution failed
+                if not team_result.get("success"):
+                    self._state.status = "failed"
+                    self._state.error_message = team_result.get("error")
+                    break
+
+                # Update execution status to waiting_for_input
+                self._state.status = "waiting_for_input"
+                self._state.is_waiting_for_input = True
+
+                # Update database to reflect waiting state
+                await workflow.execute_activity(
+                    update_execution_status,
+                    ActivityUpdateExecutionInput(
+                        execution_id=input.execution_id,
+                        status="waiting_for_input",
+                        response=self._state.current_response,
+                        usage=self._state.usage,
+                        execution_metadata={
+                            **self._state.metadata,
+                            "conversation_turns": conversation_turn,
+                            "waiting_for_user": True,
+                        },
+                    ),
+                    start_to_close_timeout=timedelta(seconds=30),
+                )
+
+                workflow.logger.info(
+                    f"Waiting for user input after team turn {conversation_turn}",
+                    extra={"turn": conversation_turn}
+                )
+
+                # Wait for either new message or mark as done
+                await workflow.wait_condition(
+                    lambda: self._new_message_count > self._processed_message_count or self._state.should_complete,
+                    timeout=timedelta(hours=24)
+                )
+
+                if self._state.should_complete:
+                    workflow.logger.info("User marked team workflow as done")
+                    break
+
+                # Continue loop to process new message
+                self._state.status = "running"
+
+            # Conversation complete - finalize workflow
+            final_status = "failed" if self._state.status == "failed" else "completed"
+            self._state.status = final_status
+
+            await workflow.execute_activity(
+                update_execution_status,
+                ActivityUpdateExecutionInput(
+                    execution_id=input.execution_id,
+                    status=final_status,
+                    completed_at=workflow.now().isoformat(),
+                    response=self._state.current_response,
+                    error_message=self._state.error_message,
+                    usage=self._state.usage,
+                    execution_metadata={
+                        **self._state.metadata,
+                        "workflow_completed": True,
+                        "total_conversation_turns": conversation_turn,
+                    },
+                ),
+                start_to_close_timeout=timedelta(seconds=30),
+            )
+
+            workflow.logger.info(
+                f"Team execution workflow completed with HITL",
+                extra={
+                    "execution_id": input.execution_id,
+                    "status": final_status,
+                    "conversation_turns": conversation_turn,
+                }
+            )
+
+            return {
+                "success": final_status == "completed",
+                "execution_id": input.execution_id,
+                "status": final_status,
+                "response": self._state.current_response,
+                "usage": self._state.usage,
+                "conversation_turns": conversation_turn,
+            }
+
+        except Exception as e:
+            # Update state with error
+            self._state.status = "failed"
+            self._state.error_message = str(e)
+            self._state.metadata["error_type"] = type(e).__name__
+
+            workflow.logger.error(
+                f"Team execution workflow failed",
+                extra={
+                    "execution_id": input.execution_id,
+                    "error": str(e),
+                }
+            )
+
+            # Update execution as failed
+            try:
+                await workflow.execute_activity(
+                    update_execution_status,
+                    ActivityUpdateExecutionInput(
+                        execution_id=input.execution_id,
+                        status="failed",
+                        completed_at=workflow.now().isoformat(),
+                        error_message=f"Workflow error: {str(e)}",
+                        execution_metadata={
+                            "workflow_error": True,
+                            "error_type": type(e).__name__,
+                        },
+                    ),
+                    start_to_close_timeout=timedelta(seconds=30),
+                )
+            except Exception as update_error:
+                workflow.logger.error(
+                    f"Failed to update status after error",
+                    extra={"error": str(update_error)}
+                )
+
+            raise

control_plane_api/scripts/seed_models.py

@@ -0,0 +1,239 @@
+"""
+Seed script for populating LLM models in the database.
+
+Run this script to populate the database with default Kubiya-supported models.
+
+Usage:
+    python -m control_plane_api.scripts.seed_models
+"""
+import sys
+from pathlib import Path
+
+# Add parent directory to path
+sys.path.insert(0, str(Path(__file__).parent.parent.parent))
+
+from control_plane_api.app.database import SessionLocal
+from control_plane_api.app.models.llm_model import LLMModel
+from datetime import datetime
+
+
+# Default models to seed
+DEFAULT_MODELS = [
+    {
+        "value": "kubiya/claude-sonnet-4",
+        "label": "Claude Sonnet 4",
+        "provider": "Anthropic",
+        "logo": "/logos/claude-color.svg",
+        "description": "Most intelligent model with best reasoning capabilities",
+        "enabled": True,
+        "recommended": True,
+        "compatible_runtimes": ["default", "claude_code"],
+        "capabilities": {
+            "vision": False,
+            "function_calling": True,
+            "max_tokens": 8192,
+            "context_window": 200000,
+        },
+        "display_order": 1,
+    },
+    {
+        "value": "kubiya/claude-opus-4",
+        "label": "Claude Opus 4",
+        "provider": "Anthropic",
+        "logo": "/logos/claude-color.svg",
+        "description": "Powerful model for complex tasks requiring deep analysis",
+        "enabled": True,
+        "recommended": False,
+        "compatible_runtimes": ["default", "claude_code"],
+        "capabilities": {
+            "vision": True,
+            "function_calling": True,
+            "max_tokens": 4096,
+            "context_window": 200000,
+        },
+        "display_order": 2,
+    },
+    {
+        "value": "kubiya/claude-3-5-sonnet-20241022",
+        "label": "Claude 3.5 Sonnet",
+        "provider": "Anthropic",
+        "logo": "/logos/claude-color.svg",
+        "description": "Previous generation Sonnet with excellent performance",
+        "enabled": True,
+        "recommended": False,
+        "compatible_runtimes": ["default", "claude_code"],
+        "capabilities": {
+            "vision": True,
+            "function_calling": True,
+            "max_tokens": 8192,
+            "context_window": 200000,
+        },
+        "display_order": 3,
+    },
+    {
+        "value": "kubiya/gpt-4o",
+        "label": "GPT-4o",
+        "provider": "OpenAI",
+        "logo": "/thirdparty/logos/openai.svg",
+        "description": "Fast and capable model with vision support",
+        "enabled": True,
+        "recommended": False,
+        "compatible_runtimes": ["default"],
+        "capabilities": {
+            "vision": True,
+            "function_calling": True,
+            "max_tokens": 16384,
+            "context_window": 128000,
+        },
+        "display_order": 4,
+    },
+    {
+        "value": "kubiya/gpt-4-turbo",
+        "label": "GPT-4 Turbo",
+        "provider": "OpenAI",
+        "logo": "/thirdparty/logos/openai.svg",
+        "description": "Enhanced GPT-4 with improved speed and capabilities",
+        "enabled": True,
+        "recommended": False,
+        "compatible_runtimes": ["default"],
+        "capabilities": {
+            "vision": True,
+            "function_calling": True,
+            "max_tokens": 4096,
+            "context_window": 128000,
+        },
+        "display_order": 5,
+    },
+    {
+        "value": "kubiya/gpt-4o-mini",
+        "label": "GPT-4o Mini",
+        "provider": "OpenAI",
+        "logo": "/thirdparty/logos/openai.svg",
+        "description": "Cost-effective model for simpler tasks",
+        "enabled": True,
+        "recommended": False,
+        "compatible_runtimes": ["default"],
+        "capabilities": {
+            "vision": True,
+            "function_calling": True,
+            "max_tokens": 16384,
+            "context_window": 128000,
+        },
+        "display_order": 6,
+    },
+    {
+        "value": "kubiya/gemini-pro",
+        "label": "Gemini Pro",
+        "provider": "Google",
+        "logo": "/thirdparty/logos/google.svg",
+        "description": "Google's powerful multimodal model",
+        "enabled": True,
+        "recommended": False,
+        "compatible_runtimes": ["default"],
+        "capabilities": {
+            "vision": True,
+            "function_calling": True,
+            "max_tokens": 8192,
+            "context_window": 1000000,
+        },
+        "display_order": 7,
+    },
+]
+
+
+def seed_models(force: bool = False):
+    """
+    Seed the database with default models.
+
+    Args:
+        force: If True, update existing models. If False, skip existing models.
+    """
+    db = SessionLocal()
+    try:
+        created_count = 0
+        updated_count = 0
+        skipped_count = 0
+
+        for model_data in DEFAULT_MODELS:
+            # Check if model exists
+            existing = db.query(LLMModel).filter(LLMModel.value == model_data["value"]).first()
+
+            if existing:
+                if force:
+                    # Update existing model
+                    for key, value in model_data.items():
+                        setattr(existing, key, value)
+                    existing.updated_at = datetime.utcnow()
+                    print(f"✓ Updated: {model_data['label']} ({model_data['value']})")
+                    updated_count += 1
+                else:
+                    print(f"⊙ Skipped (exists): {model_data['label']} ({model_data['value']})")
+                    skipped_count += 1
+            else:
+                # Create new model
+                new_model = LLMModel(**model_data)
+                db.add(new_model)
+                print(f"✓ Created: {model_data['label']} ({model_data['value']})")
+                created_count += 1
+
+        db.commit()
+
+        print("\n" + "=" * 60)
+        print("Seeding complete!")
+        print(f"  Created: {created_count}")
+        print(f"  Updated: {updated_count}")
+        print(f"  Skipped: {skipped_count}")
+        print(f"  Total: {len(DEFAULT_MODELS)}")
+        print("=" * 60)
+
+    except Exception as e:
+        db.rollback()
+        print(f"\n❌ Error seeding models: {e}")
+        raise
+    finally:
+        db.close()
+
+
+def clear_models():
+    """Clear all models from the database (use with caution!)"""
+    db = SessionLocal()
+    try:
+        count = db.query(LLMModel).delete()
+        db.commit()
+        print(f"✓ Cleared {count} models from database")
+    except Exception as e:
+        db.rollback()
+        print(f"❌ Error clearing models: {e}")
+        raise
+    finally:
+        db.close()
+
+
+if __name__ == "__main__":
+    import argparse
+
+    parser = argparse.ArgumentParser(description="Seed LLM models database")
+    parser.add_argument(
+        "--force",
+        action="store_true",
+        help="Update existing models (default: skip existing)",
+    )
+    parser.add_argument(
+        "--clear",
+        action="store_true",
+        help="Clear all models before seeding (DANGEROUS)",
+    )
+
+    args = parser.parse_args()
+
+    if args.clear:
+        confirm = input("⚠️  This will delete ALL models. Are you sure? (yes/no): ")
+        if confirm.lower() == "yes":
+            clear_models()
+        else:
+            print("Cancelled.")
+            sys.exit(0)
+
+    print("Seeding models...")
+    print("=" * 60)
+    seed_models(force=args.force)