lucidicai 1.3.5__py3-none-any.whl → 2.0.1__py3-none-any.whl

lucidicai/client.py

@@ -1,22 +1,28 @@
 import os
 import time
+import threading
 from datetime import datetime, timezone
-from typing import Optional, Tuple
+from typing import Optional, Tuple, Dict, Any
 
 import requests
 import logging
+import json
 from requests.adapters import HTTPAdapter, Retry
 from urllib3.util import Retry
 
 
 from .errors import APIKeyVerificationError, InvalidOperationError, LucidicNotInitializedError
-from .telemetry.base_provider import BaseProvider 
 from .session import Session
 from .singleton import singleton, clear_singletons
 from .lru import LRUCache
+from .event import Event
+from .event_queue import EventQueue
+import uuid
 
 NETWORK_RETRIES = 3
 
+logger = logging.getLogger("Lucidic")
+
 
 @singleton
 class Client:
@@ -25,16 +31,16 @@ class Client:
         api_key: str,
         agent_id: str,
     ):
-        self.base_url = "https://analytics.lucidic.ai/api" if not (os.getenv("LUCIDIC_DEBUG", 'False') == 'True') else "http://localhost:8000/api"
+        self.base_url = "https://backend.lucidic.ai/api" if not (os.getenv("LUCIDIC_DEBUG", 'False') == 'True') else "http://localhost:8000/api"
         self.initialized = False
         self.session = None
         self.previous_sessions = LRUCache(500)  # For LRU cache of previously initialized sessions
         self.custom_session_id_translations = LRUCache(500) # For translations of custom session IDs to real session IDs
-        self.providers = []
         self.api_key = api_key
         self.agent_id = agent_id
         self.masking_function = None
         self.auto_end = False  # Default to False until explicitly set during init
+        self._shutdown = False  # Flag to prevent requests after shutdown
         self.request_session = requests.Session()
         retry_cfg = Retry(
             total=3,                     # 3 attempts in total
@@ -46,6 +52,19 @@ class Client:
         self.request_session.mount("https://", adapter)
         self.set_api_key(api_key)
         self.prompts = dict()
+        # Initialize event queue (non-blocking event delivery)
+        self._event_queue = EventQueue(self)
+        
+        # Track telemetry state to prevent re-initialization
+        # These are process-wide singletons for telemetry
+        self._telemetry_lock = threading.Lock()  # Prevent race conditions
+        self._tracer_provider = None
+        self._instrumentors = {}  # Dict to track which providers are instrumented
+        self._telemetry_initialized = False
+        
+        # Track active sessions to prevent premature EventQueue shutdown
+        self._active_sessions_lock = threading.Lock()
+        self._active_sessions = set()  # Set of active session IDs
 
     def set_api_key(self, api_key: str):
         self.api_key = api_key
@@ -56,34 +75,23 @@ class Client:
             raise APIKeyVerificationError("Invalid API Key")
 
     def clear(self):
-        self.undo_overrides()
+        # Clean up singleton state
         clear_singletons()
         self.initialized = False
         self.session = None
-        self.providers = []
         del self
 
     def verify_api_key(self, base_url: str, api_key: str) -> Tuple[str, str]:
         data = self.make_request('verifyapikey', 'GET', {})  # TODO: Verify against agent ID provided
         return data["project"], data["project_id"]
 
-    def set_provider(self, provider: BaseProvider) -> None:
-        """Set the LLM provider to track"""
-        # Avoid duplicate provider registration of the same class
-        for existing in self.providers:
-            if type(existing) is type(provider):
-                return
-        self.providers.append(provider)
-        provider.override()
-
-    def undo_overrides(self):
-        for provider in self.providers:
-            provider.undo_override()
+    def set_provider(self, provider) -> None:
+        """Deprecated: manual provider overrides removed (no-op)."""
+        return
 
     def init_session(
         self,
         session_name: str,
-        mass_sim_id: Optional[str] = None,
         task: Optional[str] = None,
         rubrics: Optional[list] = None,
         tags: Optional[list] = None,
@@ -111,7 +119,6 @@ class Client:
             "agent_id": self.agent_id,
             "session_name": session_name,
             "task": task,
-            "mass_sim_id": mass_sim_id,
             "experiment_id": experiment_id,
             "rubrics": rubrics,
             "tags": tags,
@@ -129,47 +136,47 @@ class Client:
             agent_id=self.agent_id,
             session_id=real_session_id,
             session_name=session_name,
-            mass_sim_id=mass_sim_id,
             experiment_id=experiment_id,
             task=task,
             rubrics=rubrics,
             tags=tags,
         )
+        
+        # Track this as an active session
+        with self._active_sessions_lock:
+            self._active_sessions.add(real_session_id)
+            if logger.isEnabledFor(logging.DEBUG):
+                logger.debug(f"[Client] Added active session {real_session_id[:8]}..., total: {len(self._active_sessions)}")
+        
         self.initialized = True
         return self.session.session_id
 
+    def mark_session_inactive(self, session_id: str) -> None:
+        """Mark a session as inactive. Used when ending a session."""
+        with self._active_sessions_lock:
+            if session_id in self._active_sessions:
+                self._active_sessions.discard(session_id)
+                if logger.isEnabledFor(logging.DEBUG):
+                    logger.debug(f"[Client] Removed active session {session_id[:8]}..., remaining: {len(self._active_sessions)}")
+    
+    def has_active_sessions(self) -> bool:
+        """Check if there are any active sessions."""
+        with self._active_sessions_lock:
+            return len(self._active_sessions) > 0
+    
     def create_event_for_session(self, session_id: str, **kwargs) -> str:
-        """Create an event for a specific session id without mutating global session.
+        """Create an event for a specific session id (new typed model).
 
-        This avoids cross-thread races by not switching the active session on
-        the singleton client. It constructs an ephemeral Session facade to send
-        requests under the provided session id.
+        This avoids mutating the global session and directly uses the new
+        event API. Prefer passing typed fields and a 'type' argument.
         """
-        temp_session = Session(agent_id=self.agent_id, session_id=session_id)
-        return temp_session.create_event(**kwargs)
-
-    def continue_session(self, session_id: str):
-        if session_id in self.custom_session_id_translations:
-            session_id = self.custom_session_id_translations[session_id]
-        if self.session and self.session.session_id == session_id:
-            return self.session.session_id
-        if self.session:
-            self.previous_sessions[self.session.session_id] = self.session
-        data = self.make_request('continuesession', 'POST', {"session_id": session_id})
-        real_session_id = data["session_id"]
-        if session_id != real_session_id:
-            self.custom_session_id_translations[session_id] = real_session_id
-        self.session = Session(
-            agent_id=self.agent_id,
-            session_id=real_session_id
-        )
-        import logging as _logging
-        _logging.getLogger('Lucidic').info(f"Session {data.get('session_name', '')} continuing...")
-        return self.session.session_id
+        kwargs = dict(kwargs)
+        kwargs['session_id'] = session_id
+        return self.create_event(**kwargs)
 
-    def init_mass_sim(self, **kwargs) -> str:
+    def create_experiment(self, **kwargs) -> str:
         kwargs['agent_id'] = self.agent_id
-        return self.make_request('initmasssim', 'POST', kwargs)['mass_sim_id']
+        return self.make_request('createexperiment', 'POST', kwargs)['experiment_id']
 
     def get_prompt(self, prompt_name, cache_ttl, label) -> str:
         current_time = time.time()
@@ -194,6 +201,13 @@ class Client:
         return prompt
 
     def make_request(self, endpoint, method, data):
+        # Check if client is shutting down
+        if self._shutdown:
+            logger.warning(f"[HTTP] Attempted request after shutdown: {endpoint}")
+            return {}
+
+        data = {k: v for k, v in data.items() if v is not None}
+
         http_methods = {
             "GET": lambda data: self.request_session.get(f"{self.base_url}/{endpoint}", params=data),
             "POST": lambda data: self.request_session.post(f"{self.base_url}/{endpoint}", json=data),
@@ -201,7 +215,14 @@ class Client:
             "DELETE": lambda data: self.request_session.delete(f"{self.base_url}/{endpoint}", params=data),
         }  # TODO: make into enum
         data['current_time'] = datetime.now().astimezone(timezone.utc).isoformat()
+        # Debug: print final payload about to be sent
+        try:
+            dbg = json.dumps({"endpoint": endpoint, "method": method, "body": data}, ensure_ascii=False)
+            logger.debug(f"[HTTP] Sending request: {dbg}")
+        except Exception:
+            logger.debug(f"[HTTP] Sending request to {endpoint} {method}")
         func = http_methods[method]
+        response = None
         for _ in range(NETWORK_RETRIES):
             try:
                 response = func(data)
@@ -222,6 +243,150 @@ class Client:
             raise InvalidOperationError(f"Request to Lucidic AI Backend failed: {e.response.text}")
         return response.json()
 
+    # ==== New Typed Event Model Helpers ====
+    def _build_payload(self, type: str, kwargs: Dict[str, Any]) -> Dict[str, Any]:
+        """Build type-specific payload and place unrecognized keys in misc."""
+        # Remove non-payload top-level fields from kwargs copy
+        non_payload_fields = [
+            'parent_event_id', 'tags', 'metadata', 'occurred_at', 'duration', 'session_id',
+            'event_id'
+        ]
+        for field in non_payload_fields:
+            if field in kwargs:
+                kwargs.pop(field, None)
+
+        if type == "llm_generation":
+            return self._build_llm_payload(kwargs)
+        elif type == "function_call":
+            return self._build_function_payload(kwargs)
+        elif type == "error_traceback":
+            return self._build_error_payload(kwargs)
+        else:
+            return self._build_generic_payload(kwargs)
+
+    def _build_llm_payload(self, kwargs: Dict[str, Any]) -> Dict[str, Any]:
+        payload: Dict[str, Any] = {
+            "request": {},
+            "response": {},
+            "usage": {},
+            "status": "ok",
+            "misc": {}
+        }
+        # Request fields
+        for field in ["provider", "model", "messages", "params"]:
+            if field in kwargs:
+                payload["request"][field] = kwargs.pop(field)
+        # Response fields
+        for field in ["output", "messages", "tool_calls", "thinking", "raw"]:
+            if field in kwargs:
+                payload["response"][field] = kwargs.pop(field)
+        # Usage fields
+        for field in ["input_tokens", "output_tokens", "cache", "cost"]:
+            if field in kwargs:
+                payload["usage"][field] = kwargs.pop(field)
+        # Status / error
+        if 'status' in kwargs:
+            payload['status'] = kwargs.pop('status')
+        if 'error' in kwargs:
+            payload['error'] = kwargs.pop('error')
+        payload["misc"] = kwargs
+        return payload
+
+    def _build_function_payload(self, kwargs: Dict[str, Any]) -> Dict[str, Any]:
+        payload: Dict[str, Any] = {
+            "function_name": kwargs.pop("function_name", "unknown"),
+            "arguments": kwargs.pop("arguments", {}),
+            "return_value": kwargs.pop("return_value", None),
+            "misc": kwargs
+        }
+        return payload
+
+    def _build_error_payload(self, kwargs: Dict[str, Any]) -> Dict[str, Any]:
+        payload: Dict[str, Any] = {
+            "error": kwargs.pop("error", ""),
+            "traceback": kwargs.pop("traceback", ""),
+            "misc": kwargs
+        }
+        return payload
+
+    def _build_generic_payload(self, kwargs: Dict[str, Any]) -> Dict[str, Any]:
+        payload: Dict[str, Any] = {
+            "details": kwargs.pop("details", kwargs.pop("description", "")),
+            "misc": kwargs
+        }
+        return payload
+
+    def create_event(self, type: str = "generic", **kwargs) -> str:
+        """Create a typed event (non-blocking) and return client-side UUID.
+
+        - Generates and returns client_event_id immediately
+        - Enqueues the full event for background processing via EventQueue
+        - Supports parent nesting via client-side parent_event_id
+        - Handles client-side blob thresholding in the queue
+        """
+        # Resolve session_id: explicit -> context -> current session
+        session_id = kwargs.pop('session_id', None)
+        if not session_id:
+            try:
+                from .context import current_session_id
+                session_id = current_session_id.get(None)
+            except Exception:
+                session_id = None
+        if not session_id and self.session:
+            session_id = self.session.session_id
+        if not session_id:
+            raise InvalidOperationError("No active session for event creation")
+
+        # Parent event id from kwargs or parent context (client-side)
+        parent_event_id = kwargs.get('parent_event_id')
+        if not parent_event_id:
+            try:
+                from .context import current_parent_event_id
+                parent_event_id = current_parent_event_id.get(None)
+            except Exception:
+                parent_event_id = None
+
+        # Build payload (typed)
+        payload = self._build_payload(type, dict(kwargs))
+
+        # Occurred-at
+        from datetime import datetime as _dt
+        _occ = kwargs.get("occurred_at")
+        if isinstance(_occ, str):
+            occurred_at_str = _occ
+        elif isinstance(_occ, _dt):
+            if _occ.tzinfo is None:
+                local_tz = _dt.now().astimezone().tzinfo
+                occurred_at_str = _occ.replace(tzinfo=local_tz).isoformat()
+            else:
+                occurred_at_str = _occ.isoformat()
+        else:
+            occurred_at_str = _dt.now().astimezone().isoformat()
+
+        # Client-side UUIDs
+        client_event_id = kwargs.get('event_id') or str(uuid.uuid4())
+
+        # Build request body with client ids
+        event_request: Dict[str, Any] = {
+            "session_id": session_id,
+            "client_event_id": client_event_id,
+            "client_parent_event_id": parent_event_id,
+            "type": type,
+            "occurred_at": occurred_at_str,
+            "duration": kwargs.get("duration"),
+            "tags": kwargs.get("tags", []),
+            "metadata": kwargs.get("metadata", {}),
+            "payload": payload,
+        }
+
+        # Queue for background processing and return immediately
+        self._event_queue.queue_event(event_request)
+        return client_event_id
+
+    def update_event(self, event_id: str, type: Optional[str] = None, **kwargs) -> str:
+        """Deprecated: events are immutable in the new model."""
+        raise InvalidOperationError("update_event is no longer supported. Events are immutable.")
+
     def mask(self, data):
         if not self.masking_function:
             return data
@@ -232,4 +397,114 @@ class Client:
         except Exception as e:
             logger = logging.getLogger('Lucidic')
             logger.error(f"Error in custom masking function: {repr(e)}")
-            return "<Error in custom masking function, this is a fully-masked placeholder>"
+            return "<Error in custom masking function, this is a fully-masked placeholder>"
+    
+    def initialize_telemetry(self, providers: list) -> bool:
+        """
+        Initialize telemetry with the given providers.
+        This is a true singleton - only the first call creates the TracerProvider.
+        Subsequent calls only add new instrumentors if needed.
+        
+        Args:
+            providers: List of provider names to instrument
+            
+        Returns:
+            True if telemetry was successfully initialized or already initialized
+        """
+        with self._telemetry_lock:
+            try:
+                # Create TracerProvider only once per process
+                if self._tracer_provider is None:
+                    logger.debug("[Telemetry] Creating TracerProvider (first initialization)")
+                    
+                    from opentelemetry import trace
+                    from opentelemetry.sdk.trace import TracerProvider
+                    from opentelemetry.sdk.trace.export import BatchSpanProcessor
+                    from opentelemetry.sdk.resources import Resource
+                    
+                    resource = Resource.create({
+                        "service.name": "lucidic-ai",
+                        "service.version": "1.0.0",
+                        "lucidic.agent_id": self.agent_id,
+                    })
+                    
+                    # Create provider with shutdown_on_exit=False for our control
+                    self._tracer_provider = TracerProvider(resource=resource, shutdown_on_exit=False)
+                    
+                    # Add context capture processor FIRST
+                    from .telemetry.context_capture_processor import ContextCaptureProcessor
+                    context_processor = ContextCaptureProcessor()
+                    self._tracer_provider.add_span_processor(context_processor)
+                    
+                    # Add exporter processor for sending spans to Lucidic
+                    from .telemetry.lucidic_exporter import LucidicSpanExporter
+                    exporter = LucidicSpanExporter()
+                    # Configure for faster export: 100ms interval instead of default 5000ms
+                    # This matches the TypeScript SDK's flush interval pattern
+                    export_processor = BatchSpanProcessor(
+                        exporter,
+                        schedule_delay_millis=100,  # Export every 100ms
+                        max_export_batch_size=512,  # Reasonable batch size
+                        max_queue_size=2048         # Larger queue for burst handling
+                    )
+                    self._tracer_provider.add_span_processor(export_processor)
+                    
+                    # Set as global provider (only happens once)
+                    try:
+                        trace.set_tracer_provider(self._tracer_provider)
+                        logger.debug("[Telemetry] Set global TracerProvider")
+                    except Exception as e:
+                        # This is OK - might already be set
+                        logger.debug(f"[Telemetry] Global provider already set: {e}")
+                    
+                    self._telemetry_initialized = True
+                
+                # Now instrument the requested providers (can happen multiple times)
+                if providers:
+                    from .telemetry.telemetry_init import instrument_providers
+                    new_instrumentors = instrument_providers(providers, self._tracer_provider, self._instrumentors)
+                    # Update our tracking dict
+                    self._instrumentors.update(new_instrumentors)
+                    logger.debug(f"[Telemetry] Instrumented providers: {list(new_instrumentors.keys())}")
+                
+                return True
+                
+            except Exception as e:
+                logger.error(f"[Telemetry] Failed to initialize: {e}")
+                return False
+    
+    def flush_telemetry(self, timeout_seconds: float = 2.0) -> bool:
+        """
+        Flush all OpenTelemetry spans to ensure they're exported.
+        
+        This method blocks until all buffered spans in the TracerProvider
+        are exported or the timeout is reached. Critical for ensuring
+        LLM generation events are not lost during shutdown.
+        
+        Handles both active and shutdown TracerProviders gracefully.
+        
+        Args:
+            timeout_seconds: Maximum time to wait for flush completion
+            
+        Returns:
+            True if flush succeeded, False if timeout occurred
+        """
+        try:
+            if self._tracer_provider:
+                # Check if provider is already shutdown
+                if hasattr(self._tracer_provider, '_shutdown') and self._tracer_provider._shutdown:
+                    logger.debug("[Telemetry] TracerProvider already shutdown, skipping flush")
+                    return True
+                    
+                # Convert seconds to milliseconds for OpenTelemetry
+                timeout_millis = int(timeout_seconds * 1000)
+                success = self._tracer_provider.force_flush(timeout_millis)
+                if success:
+                    logger.debug(f"[Telemetry] Successfully flushed spans (timeout={timeout_seconds}s)")
+                else:
+                    logger.warning(f"[Telemetry] Flush timed out after {timeout_seconds}s")
+                return success
+            return True  # No provider = nothing to flush = success
+        except Exception as e:
+            logger.error(f"[Telemetry] Failed to flush spans: {e}")
+            return False

lucidicai/constants.py

@@ -1,33 +1,6 @@
-"""Constants used throughout the Lucidic SDK"""
+"""Constants used throughout the Lucidic SDK (steps removed)."""
 
-# Step states
-class StepState:
-    """Constants for step states"""
-    RUNNING = "Running: {agent_name}"
-    FINISHED = "Finished: {agent_name}"
-    HANDOFF = "Handoff: {agent_name}"
-    TRANSFERRED = "Transferred to {agent_name}"
-    ERROR = "Error in {agent_name}"
-
-# Step actions  
-class StepAction:
-    """Constants for step actions"""
-    EXECUTE = "Execute {agent_name}"
-    TRANSFER = "Transfer from {from_agent}"
-    HANDOFF = "Handoff from {from_agent}"
-    DELIVERED = "{agent_name} finished processing"
-    FAILED = "Agent execution failed"
-
-# Step goals
-class StepGoal:
-    """Constants for step goals"""
-    PROCESS_REQUEST = "Process request"
-    CONTINUE_PROCESSING = "Continue processing"
-    CONTINUE_WITH = "Continue with {agent_name}"
-    PROCESSING_FINISHED = "Processing finished"
-    ERROR = "Error: {error}"
-
-# Event descriptions
+# Event descriptions (generic)
 class EventDescription:
     """Constants for event descriptions"""
     TOOL_CALL = "Tool call: {tool_name}"
@@ -48,12 +21,9 @@ class LogMessage:
     """Constants for log messages"""
     SESSION_INIT = "Session initialized successfully"
     SESSION_CONTINUE = "Session {session_id} continuing..."
-    INSTRUMENTATION_ENABLED = "OpenAI Agents SDK instrumentation enabled"
-    INSTRUMENTATION_DISABLED = "OpenAI Agents SDK instrumentation disabled"
-    NO_ACTIVE_SESSION = "No active session for agent tracking"
+    INSTRUMENTATION_ENABLED = "Instrumentation enabled"
+    INSTRUMENTATION_DISABLED = "Instrumentation disabled"
+    NO_ACTIVE_SESSION = "No active session for tracking"
     HANDLER_INTERCEPTED = "Intercepted {method} call"
-    AGENT_RUNNING = "Running agent '{agent_name}' with prompt: {prompt}"
-    AGENT_COMPLETED = "Agent completed successfully"
-    STEP_CREATED = "Created step: {step_id}"
-    STEP_ENDED = "Step ended: {step_id}"
-    HANDOFF_DETECTED = "Handoff chain detected: {chain}"
+    AGENT_RUNNING = "Running agent '{agent_name}'"
+    AGENT_COMPLETED = "Agent completed successfully"

lucidicai/context.py

@@ -19,6 +19,12 @@ current_session_id: contextvars.ContextVar[Optional[str]] = contextvars.ContextV
 )
 
 
+# NEW: Context variable for parent event nesting
+current_parent_event_id: contextvars.ContextVar[Optional[str]] = contextvars.ContextVar(
+    "lucidic.parent_event_id", default=None
+)
+
+
 def set_active_session(session_id: Optional[str]) -> None:
     """Bind the given session id to the current execution context."""
     current_session_id.set(session_id)
@@ -49,6 +55,25 @@ async def bind_session_async(session_id: str) -> AsyncIterator[None]:
         current_session_id.reset(token)
 
 
+# NEW: Parent event context managers
+@contextmanager
+def event_context(event_id: str) -> Iterator[None]:
+    token = current_parent_event_id.set(event_id)
+    try:
+        yield
+    finally:
+        current_parent_event_id.reset(token)
+
+
+@asynccontextmanager
+async def event_context_async(event_id: str) -> AsyncIterator[None]:
+    token = current_parent_event_id.set(event_id)
+    try:
+        yield
+    finally:
+        current_parent_event_id.reset(token)
+
+
 @contextmanager
 def session(**init_params) -> Iterator[None]:
     """All-in-one context manager: init → bind → yield → clear → end.