letta-server/letta/server/rest_api/app.py

import faulthandler
import importlib.util
import json
import logging
import os
import platform
import sys
import threading
from contextlib import asynccontextmanager
from functools import partial
from pathlib import Path
from typing import Optional

import uvicorn

# Enable Python fault handler to get stack traces on segfaults
faulthandler.enable()

from fastapi import FastAPI, Request
from fastapi.exceptions import RequestValidationError
from fastapi.responses import JSONResponse, ORJSONResponse
from marshmallow import ValidationError
from sqlalchemy.exc import IntegrityError, OperationalError
from starlette.middleware.cors import CORSMiddleware

from letta.__init__ import __version__ as letta_version
from letta.agents.exceptions import IncompatibleAgentType
from letta.constants import ADMIN_PREFIX, API_PREFIX, OPENAI_API_PREFIX
from letta.errors import (
    AgentExportIdMappingError,
    AgentExportProcessingError,
    AgentFileImportError,
    AgentNotFoundForExportError,
    BedrockPermissionError,
    ConcurrentUpdateError,
    ConversationBusyError,
    EmbeddingConfigRequiredError,
    HandleNotFoundError,
    LettaAgentNotFoundError,
    LettaExpiredError,
    LettaImageFetchError,
    LettaInvalidArgumentError,
    LettaInvalidMCPSchemaError,
    LettaMCPConnectionError,
    LettaMCPTimeoutError,
    LettaServiceUnavailableError,
    LettaToolCreateError,
    LettaToolNameConflictError,
    LettaUnsupportedFileUploadError,
    LettaUserNotFoundError,
    LLMAuthenticationError,
    LLMError,
    LLMProviderOverloaded,
    LLMRateLimitError,
    LLMTimeoutError,
    NoActiveRunsToCancelError,
    PendingApprovalError,
)
from letta.helpers.pinecone_utils import get_pinecone_indices, should_use_pinecone, upsert_pinecone_indices
from letta.jobs.scheduler import start_scheduler_with_leader_election
from letta.log import get_logger
from letta.orm.errors import DatabaseTimeoutError, ForeignKeyConstraintViolationError, NoResultFound, UniqueConstraintViolationError
from letta.otel.tracing import get_trace_id
from letta.schemas.letta_message import create_letta_error_message_schema, create_letta_message_union_schema
from letta.schemas.letta_message_content import (
    create_letta_assistant_message_content_union_schema,
    create_letta_message_content_union_schema,
    create_letta_tool_return_content_union_schema,
    create_letta_user_message_content_union_schema,
)
from letta.server.constants import REST_DEFAULT_PORT
from letta.server.db import db_registry
from letta.server.global_exception_handler import setup_global_exception_handlers

# NOTE(charles): these are extra routes that are not part of v1 but we still need to mount to pass tests
from letta.server.rest_api.auth.index import setup_auth_router  # TODO: probably remove right?
from letta.server.rest_api.interface import StreamingServerInterface
from letta.server.rest_api.middleware import CheckPasswordMiddleware, LoggingMiddleware, RequestIdMiddleware
from letta.server.rest_api.routers.v1 import ROUTERS as v1_routes
from letta.server.rest_api.routers.v1.organizations import router as organizations_router
from letta.server.rest_api.routers.v1.users import router as users_router  # TODO: decide on admin
from letta.server.rest_api.static_files import mount_static_files
from letta.server.rest_api.utils import SENTRY_ENABLED
from letta.server.server import SyncServer
from letta.settings import settings, telemetry_settings
from letta.validators import PATH_VALIDATORS, PRIMITIVE_ID_PATTERNS

if SENTRY_ENABLED:
    import sentry_sdk

IS_WINDOWS = platform.system() == "Windows"

# NOTE(charles): @ethan I had to add this to get the global as the bottom to work
interface: type = StreamingServerInterface
server = SyncServer(default_interface_factory=lambda: interface())
logger = get_logger(__name__)


def generate_openapi_schema(app: FastAPI):
    # Update the OpenAPI schema
    if not app.openapi_schema:
        app.openapi_schema = app.openapi()

    letta_docs = app.openapi_schema.copy()
    letta_docs["paths"] = {k: v for k, v in letta_docs["paths"].items() if not k.startswith("/openai")}
    letta_docs["info"]["title"] = "Letta API"
    letta_docs["components"]["schemas"]["LettaMessageUnion"] = create_letta_message_union_schema()
    letta_docs["components"]["schemas"]["LettaMessageContentUnion"] = create_letta_message_content_union_schema()
    letta_docs["components"]["schemas"]["LettaAssistantMessageContentUnion"] = create_letta_assistant_message_content_union_schema()
    letta_docs["components"]["schemas"]["LettaToolReturnContentUnion"] = create_letta_tool_return_content_union_schema()
    letta_docs["components"]["schemas"]["LettaUserMessageContentUnion"] = create_letta_user_message_content_union_schema()
    letta_docs["components"]["schemas"]["LettaErrorMessage"] = create_letta_error_message_schema()

    # Update the app's schema with our modified version
    app.openapi_schema = letta_docs

    for name, docs in [
        (
            "letta",
            letta_docs,
        ),
    ]:
        if settings.cors_origins:
            docs["servers"] = [{"url": host} for host in settings.cors_origins]
        Path(f"openapi_{name}.json").write_text(json.dumps(docs, indent=2))


# middleware that only allows requests to pass through if user provides a password thats randomly generated and stored in memory
def generate_password():
    import secrets

    return secrets.token_urlsafe(16)


random_password = os.getenv("LETTA_SERVER_PASSWORD") or generate_password()


@asynccontextmanager
async def lifespan(app_: FastAPI):
    """
    FastAPI lifespan context manager with setup before the app starts pre-yield and on shutdown after the yield.
    """
    worker_id = os.getpid()

    # Initialize event loop watchdog
    try:
        import asyncio

        from letta.monitoring.event_loop_watchdog import start_watchdog

        loop = asyncio.get_running_loop()
        start_watchdog(loop, check_interval=5.0, timeout_threshold=15.0)
        logger.info(f"[Worker {worker_id}] Event loop watchdog started")
    except Exception as e:
        logger.warning(f"[Worker {worker_id}] Failed to start watchdog: {e}")

    # Pre-download NLTK data to avoid blocking during requests (fallback if Docker build failed)
    try:
        import asyncio

        import nltk

        logger.info(f"[Worker {worker_id}] Checking NLTK data availability...")
        await asyncio.to_thread(nltk.download, "punkt_tab", quiet=True)
        logger.info(f"[Worker {worker_id}] NLTK data ready")
    except Exception as e:
        logger.warning(f"[Worker {worker_id}] Failed to download NLTK data: {e}")

    # Log effective database timeout settings for debugging
    try:
        from sqlalchemy import text

        from letta.server.db import db_registry

        async with db_registry.async_session() as session:
            result = await session.execute(text("SHOW statement_timeout"))
            statement_timeout = result.scalar()
            logger.warning(f"[Worker {worker_id}] PostgreSQL statement_timeout: {statement_timeout}")
    except Exception as e:
        logger.warning(f"[Worker {worker_id}] Failed to query statement_timeout: {e}")

    if should_use_pinecone():
        if settings.upsert_pinecone_indices:
            logger.info(f"[Worker {worker_id}] Upserting pinecone indices: {get_pinecone_indices()}")
            await upsert_pinecone_indices()
            logger.info(f"[Worker {worker_id}] Upserted pinecone indices")
        else:
            logger.info(f"[Worker {worker_id}] Enabled pinecone")
    else:
        logger.info(f"[Worker {worker_id}] Disabled pinecone")

    logger.info(f"[Worker {worker_id}] Starting scheduler with leader election")
    global server
    await server.init_async(init_with_default_org_and_user=not settings.no_default_actor)
    try:
        await start_scheduler_with_leader_election(server)
        logger.info(f"[Worker {worker_id}] Scheduler initialization completed")
    except Exception as e:
        logger.error(f"[Worker {worker_id}] Scheduler initialization failed: {e}", exc_info=True)
    logger.info(f"[Worker {worker_id}] Lifespan startup completed")
    yield

    # Cleanup on shutdown
    logger.info(f"[Worker {worker_id}] Starting lifespan shutdown")

    try:
        from letta.jobs.scheduler import shutdown_scheduler_and_release_lock

        await shutdown_scheduler_and_release_lock()
        logger.info(f"[Worker {worker_id}] Scheduler shutdown completed")
    except Exception as e:
        logger.error(f"[Worker {worker_id}] Scheduler shutdown failed: {e}", exc_info=True)

    # Cleanup SQLAlchemy instrumentation
    if not settings.disable_tracing and settings.sqlalchemy_tracing:
        try:
            from letta.otel.sqlalchemy_instrumentation_integration import teardown_letta_db_instrumentation

            teardown_letta_db_instrumentation()
            logger.info(f"[Worker {worker_id}] SQLAlchemy instrumentation shutdown completed")
        except Exception as e:
            logger.warning(f"[Worker {worker_id}] SQLAlchemy instrumentation shutdown failed: {e}")

    # Shutdown LLM raw trace writer (closes ClickHouse connection)
    if settings.store_llm_traces:
        try:
            from letta.services.llm_trace_writer import get_llm_trace_writer

            writer = get_llm_trace_writer()
            await writer.shutdown_async()
            logger.info(f"[Worker {worker_id}] LLM raw trace writer shutdown completed")
        except Exception as e:
            logger.warning(f"[Worker {worker_id}] LLM raw trace writer shutdown failed: {e}")

    logger.info(f"[Worker {worker_id}] Lifespan shutdown completed")


def create_application() -> "FastAPI":
    """the application start routine"""
    # global server
    # server = SyncServer(default_interface_factory=lambda: interface())
    print(f"\n[[ Letta server // v{letta_version} ]]")

    if SENTRY_ENABLED:
        sentry_sdk.init(
            dsn=os.getenv("SENTRY_DSN"),
            environment=os.getenv("LETTA_ENVIRONMENT", "undefined"),
            traces_sample_rate=1.0,
            _experiments={
                "continuous_profiling_auto_start": True,
            },
        )

    if telemetry_settings.enable_datadog:
        try:
            dd_env = settings.environment or "development"
            print(f"▶ Initializing Datadog tracing (env={dd_env})")

            # Configure environment variables before importing ddtrace (must be set in environment before importing ddtrace)
            os.environ.setdefault("DD_ENV", dd_env)
            os.environ.setdefault("DD_SERVICE", telemetry_settings.datadog_service_name)
            os.environ.setdefault("DD_VERSION", letta_version)
            os.environ.setdefault("DD_AGENT_HOST", telemetry_settings.datadog_agent_host)
            os.environ.setdefault("DD_TRACE_AGENT_PORT", str(telemetry_settings.datadog_agent_port))
            os.environ.setdefault("DD_PROFILING_ENABLED", str(telemetry_settings.datadog_profiling_enabled).lower())
            os.environ.setdefault("DD_PROFILING_MEMORY_ENABLED", str(telemetry_settings.datadog_profiling_memory_enabled).lower())
            os.environ.setdefault("DD_PROFILING_HEAP_ENABLED", str(telemetry_settings.datadog_profiling_heap_enabled).lower())

            # Note: DD_LOGS_INJECTION, DD_APPSEC_ENABLED, DD_IAST_ENABLED, DD_APPSEC_SCA_ENABLED
            # are set via deployment configs and automatically picked up by ddtrace

            # Initialize Datadog tracer for APM (distributed tracing)
            import ddtrace

            ddtrace.patch_all()  # Auto-instrument FastAPI, HTTP, DB, etc.

            llmobs_flag = os.getenv("DD_LLMOBS_ENABLED", "")
            from ddtrace.llmobs import LLMObs

            try:
                from ddtrace.llmobs._constants import MODEL_PROVIDER
                from ddtrace.llmobs._integrations.openai import OpenAIIntegration

                if not getattr(OpenAIIntegration, "_letta_provider_patch_done", False):
                    original_set_tags = OpenAIIntegration._llmobs_set_tags

                    def _letta_set_tags(self, span, args, kwargs, response=None, operation=""):
                        original_set_tags(self, span, args, kwargs, response=response, operation=operation)

                        base_url = span.get_tag("openai.api_base")
                        if not base_url:
                            try:
                                client = getattr(self, "_client", None)
                                base_url = str(getattr(client, "_base_url", "") or "")
                            except Exception:
                                base_url = ""

                        u = (base_url or "").lower()
                        provider = None
                        if "openrouter" in u:
                            provider = "openrouter"
                        elif "groq" in u:
                            provider = "groq"

                        if provider:
                            span._set_ctx_item(MODEL_PROVIDER, provider)
                            span._set_tag_str("openai.request.provider", provider)

                    OpenAIIntegration._llmobs_set_tags = _letta_set_tags
                    OpenAIIntegration._letta_provider_patch_done = True
            except Exception:
                logger.exception("Failed to patch ddtrace OpenAI LLMObs provider detection")

            if llmobs_flag:
                LLMObs.enable(
                    ml_app=os.getenv("DD_LLMOBS_ML_APP") or telemetry_settings.datadog_service_name,
                )

            logger.info(
                f"Datadog tracer initialized: env={dd_env}, "
                f"service={telemetry_settings.datadog_service_name}, "
                f"agent={telemetry_settings.datadog_agent_host}:{telemetry_settings.datadog_agent_port}"
            )

            if telemetry_settings.datadog_profiling_enabled:
                from ddtrace.profiling import Profiler

                # Initialize and start profiler
                profiler = Profiler(
                    env=dd_env,
                    service=telemetry_settings.datadog_service_name,
                    version=letta_version,
                )
                profiler.start()

                # Log Git metadata for source code integration
                git_info = ""
                if telemetry_settings.datadog_git_commit_sha:
                    git_info = f", commit={telemetry_settings.datadog_git_commit_sha[:8]}"
                if telemetry_settings.datadog_git_repository_url:
                    git_info += f", repo={telemetry_settings.datadog_git_repository_url}"

                logger.info(
                    f"Datadog profiling enabled: env={dd_env}, "
                    f"service={telemetry_settings.datadog_service_name}, "
                    f"agent={telemetry_settings.datadog_agent_host}:{telemetry_settings.datadog_agent_port}{git_info}"
                )
        except Exception as e:
            logger.error(f"Failed to initialize Datadog tracing/profiling: {e}", exc_info=True)
            if SENTRY_ENABLED:
                sentry_sdk.capture_exception(e)
            # Don't fail application startup if Datadog initialization fails

    debug_mode = "--debug" in sys.argv
    app = FastAPI(
        swagger_ui_parameters={"docExpansion": "none"},
        # openapi_tags=TAGS_METADATA,
        title="Letta",
        summary="Create LLM agents with long-term memory and custom tools 📚🦙",
        version=letta_version,
        debug=debug_mode,  # if True, the stack trace will be printed in the response
        lifespan=lifespan,
        default_response_class=ORJSONResponse,  # Use orjson for 10x faster JSON serialization
    )

    # === Global Exception Handlers ===
    # Set up handlers for exceptions outside of request context (background tasks, threads, etc.)
    setup_global_exception_handlers()

    # === Exception Handlers ===
    # TODO (cliandy): move to separate file

    @app.exception_handler(Exception)
    async def generic_error_handler(request: Request, exc: Exception):
        # Log with structured context
        request_context = {
            "method": request.method,
            "url": str(request.url),
            "path": request.url.path,
        }

        # Extract user context if available
        user_context = {}
        if hasattr(request.state, "user_id"):
            user_context["user_id"] = request.state.user_id
        if hasattr(request.state, "org_id"):
            user_context["org_id"] = request.state.org_id

        logger.error(
            f"Unhandled error: {exc.__class__.__name__}: {str(exc)}",
            extra={
                "exception_type": exc.__class__.__name__,
                "exception_message": str(exc),
                "exception_module": exc.__class__.__module__,
                "request": request_context,
                "user": user_context,
            },
            exc_info=True,
        )

        if SENTRY_ENABLED:
            sentry_sdk.capture_exception(exc)

        return JSONResponse(
            status_code=500,
            content={
                "detail": "An unknown error occurred",
                # Only include error details in debug/development mode
                # "debug_info": str(exc) if settings.debug else None
            },
        )

    # Reasoning for this handler is the default path validation logic returns a pretty gnarly error message
    # because of the uuid4 pattern. This handler rewrites the error message to be more user-friendly and less intimidating.
    @app.exception_handler(RequestValidationError)
    async def custom_request_validation_handler(request: Request, exc: RequestValidationError):
        """Generalize path `_id` validation messages and include example IDs.

        - Rewrites string pattern/length mismatches to "primitive-{uuid4}"
        - Preserves stringified `detail` and includes `trace_id`
        - Adds top-level `examples` from `PATH_VALIDATORS` for offending params
        """
        errors = exc.errors()
        examples_set: set[str] = set()
        content = {"trace_id": get_trace_id() or ""}
        for err in errors:
            fastapi_error_loc = err.get("loc", [])
            # only rewrite path param validation errors (should expand in future)
            if len(fastapi_error_loc) != 2 or fastapi_error_loc[0] != "path":
                continue

            # re-write the error message
            parameter_name = fastapi_error_loc[1]
            err_type = err.get("type")
            if (
                err_type in {"string_pattern_mismatch", "string_too_short", "string_too_long"}
                and isinstance(parameter_name, str)
                and parameter_name.endswith("_id")
            ):
                primitive = parameter_name[:-3]
                validator = PATH_VALIDATORS.get(primitive)
                if validator:
                    # simplify default error message
                    err["msg"] = f"String should match pattern '{primitive}-{{uuid4}}'"

                    # rewrite as string_pattern_mismatch even if the input length is too short or too long (more intuitive for user)
                    if err_type in {"string_too_short", "string_too_long"}:
                        # FYI: the pattern is the same as the pattern inthe validator object but for some reason the validator object
                        # doesn't let you access it directly (unless you call into pydantic layer)
                        err["ctx"] = {"pattern": PRIMITIVE_ID_PATTERNS[primitive].pattern}
                    err["type"] = "string_pattern_mismatch"

                    # collect examples for top-level examples field (prevents duplicates and allows for multiple examples for multiple primitives)
                    # e.g. if there are 2 malformed agent ids, the examples field will contain 2 examples for the agent primitive
                    # e.g. if there is a malformed agent id and malformed folder id, the examples field will contain both examples, for both the agent and folder primitives
                    try:
                        exs = getattr(validator, "examples", None)
                        if exs:
                            for ex in exs:
                                examples_set.add(ex)
                        else:
                            examples_set.add(f"{primitive}-123e4567-e89b-42d3-8456-426614174000")
                    except Exception:
                        examples_set.add(f"{primitive}-123e4567-e89b-42d3-8456-426614174000")

        # Preserve current API contract: stringified list of errors
        content["detail"] = repr(errors)
        if examples_set:
            content["examples"] = sorted(examples_set)
        return JSONResponse(status_code=422, content=content)

    async def error_handler_with_code(request: Request, exc: Exception, code: int, detail: str | None = None):
        logger.error(f"{type(exc).__name__}", exc_info=exc)

        if not detail:
            detail = str(exc)
        return JSONResponse(
            status_code=code,
            content={"detail": detail},
        )

    _error_handler_400 = partial(error_handler_with_code, code=400)
    _error_handler_404 = partial(error_handler_with_code, code=404)
    _error_handler_404_agent = partial(_error_handler_404, detail="Agent not found")
    _error_handler_404_user = partial(_error_handler_404, detail="User not found")
    _error_handler_408 = partial(error_handler_with_code, code=408)
    _error_handler_409 = partial(error_handler_with_code, code=409)
    _error_handler_410 = partial(error_handler_with_code, code=410)
    _error_handler_415 = partial(error_handler_with_code, code=415)
    _error_handler_422 = partial(error_handler_with_code, code=422)
    _error_handler_500 = partial(error_handler_with_code, code=500)
    _error_handler_503 = partial(error_handler_with_code, code=503)

    # 400 Bad Request errors
    app.add_exception_handler(LettaInvalidArgumentError, _error_handler_400)
    app.add_exception_handler(LettaToolCreateError, _error_handler_400)
    app.add_exception_handler(LettaToolNameConflictError, _error_handler_400)
    app.add_exception_handler(AgentFileImportError, _error_handler_400)
    app.add_exception_handler(EmbeddingConfigRequiredError, _error_handler_400)
    app.add_exception_handler(LettaImageFetchError, _error_handler_400)
    app.add_exception_handler(ValueError, _error_handler_400)

    # 404 Not Found errors
    app.add_exception_handler(NoResultFound, _error_handler_404)
    app.add_exception_handler(LettaAgentNotFoundError, _error_handler_404_agent)
    app.add_exception_handler(LettaUserNotFoundError, _error_handler_404_user)
    app.add_exception_handler(AgentNotFoundForExportError, _error_handler_404)
    app.add_exception_handler(HandleNotFoundError, _error_handler_404)

    # 410 Expired errors
    app.add_exception_handler(LettaExpiredError, _error_handler_410)

    # 408 Timeout errors
    app.add_exception_handler(LettaMCPTimeoutError, _error_handler_408)
    app.add_exception_handler(LettaInvalidMCPSchemaError, _error_handler_400)

    # 409 Conflict errors
    app.add_exception_handler(ForeignKeyConstraintViolationError, _error_handler_409)
    app.add_exception_handler(UniqueConstraintViolationError, _error_handler_409)
    app.add_exception_handler(IntegrityError, _error_handler_409)
    app.add_exception_handler(ConcurrentUpdateError, _error_handler_409)
    app.add_exception_handler(ConversationBusyError, _error_handler_409)
    app.add_exception_handler(PendingApprovalError, _error_handler_409)
    app.add_exception_handler(NoActiveRunsToCancelError, _error_handler_409)

    # 415 Unsupported Media Type errors
    app.add_exception_handler(LettaUnsupportedFileUploadError, _error_handler_415)

    # 422 Validation errors
    app.add_exception_handler(ValidationError, _error_handler_422)

    # 500 Internal Server errors
    app.add_exception_handler(AgentExportIdMappingError, _error_handler_500)
    app.add_exception_handler(AgentExportProcessingError, _error_handler_500)

    # 503 Service Unavailable errors
    app.add_exception_handler(OperationalError, _error_handler_503)
    app.add_exception_handler(LettaServiceUnavailableError, _error_handler_503)
    app.add_exception_handler(LLMProviderOverloaded, _error_handler_503)

    @app.exception_handler(IncompatibleAgentType)
    async def handle_incompatible_agent_type(request: Request, exc: IncompatibleAgentType):
        logger.error("Incompatible agent types. Expected: %s, Actual: %s", exc.expected_type, exc.actual_type)
        if SENTRY_ENABLED:
            sentry_sdk.capture_exception(exc)

        return JSONResponse(
            status_code=400,
            content={
                "detail": str(exc),
                "expected_type": exc.expected_type,
                "actual_type": exc.actual_type,
            },
        )

    @app.exception_handler(DatabaseTimeoutError)
    async def database_timeout_error_handler(request: Request, exc: DatabaseTimeoutError):
        logger.error(f"Timeout occurred: {exc}. Original exception: {exc.original_exception}")
        if SENTRY_ENABLED:
            sentry_sdk.capture_exception(exc)

        return JSONResponse(
            status_code=503,
            content={"detail": "The database is temporarily unavailable. Please try again later."},
        )

    @app.exception_handler(BedrockPermissionError)
    async def bedrock_permission_error_handler(request, exc: BedrockPermissionError):
        logger.error("Bedrock permission denied.")

        return JSONResponse(
            status_code=403,
            content={
                "error": {
                    "type": "bedrock_permission_denied",
                    "message": "Unable to access the required AI model. Please check your Bedrock permissions or contact support.",
                    "detail": {str(exc)},
                }
            },
        )

    @app.exception_handler(LLMTimeoutError)
    async def llm_timeout_error_handler(request: Request, exc: LLMTimeoutError):
        return JSONResponse(
            status_code=504,
            content={
                "error": {
                    "type": "llm_timeout",
                    "message": "The LLM request timed out. Please try again.",
                    "detail": str(exc),
                }
            },
        )

    @app.exception_handler(LLMRateLimitError)
    async def llm_rate_limit_error_handler(request: Request, exc: LLMRateLimitError):
        return JSONResponse(
            status_code=429,
            content={
                "error": {
                    "type": "llm_rate_limit",
                    "message": "Rate limit exceeded for LLM model provider. Please wait before making another request.",
                    "detail": str(exc),
                }
            },
        )

    @app.exception_handler(LLMAuthenticationError)
    async def llm_auth_error_handler(request: Request, exc: LLMAuthenticationError):
        return JSONResponse(
            status_code=401,
            content={
                "error": {
                    "type": "llm_authentication",
                    "message": "Authentication failed with the LLM model provider.",
                    "detail": str(exc),
                }
            },
        )

    @app.exception_handler(LettaMCPConnectionError)
    async def mcp_connection_error_handler(request: Request, exc: LettaMCPConnectionError):
        return JSONResponse(
            status_code=502,
            content={
                "error": {
                    "type": "mcp_connection_error",
                    "message": "Failed to connect to MCP server.",
                    "detail": str(exc),
                }
            },
        )

    @app.exception_handler(LLMError)
    async def llm_error_handler(request: Request, exc: LLMError):
        return JSONResponse(
            status_code=502,
            content={
                "error": {
                    "type": "llm_error",
                    "message": "An error occurred with the LLM request.",
                    "detail": str(exc),
                }
            },
        )

    settings.cors_origins.append("https://app.letta.com")

    if (os.getenv("LETTA_SERVER_SECURE") == "true") or "--secure" in sys.argv:
        print(f"▶ Using secure mode with password: {random_password}")
        app.add_middleware(CheckPasswordMiddleware, password=random_password)

    # Add reverse proxy middleware to handle X-Forwarded-* headers
    # app.add_middleware(ReverseProxyMiddleware, base_path=settings.server_base_path)

    # Add unified logging middleware - enriches log context and logs exceptions
    app.add_middleware(LoggingMiddleware)

    # Add request ID middleware - extracts x-api-request-log-id header and sets it in contextvar
    # This is a pure ASGI middleware to properly propagate contextvars to streaming responses
    app.add_middleware(RequestIdMiddleware)

    app.add_middleware(
        CORSMiddleware,
        allow_origins=settings.cors_origins,
        allow_credentials=True,
        allow_methods=["*"],
        allow_headers=["*"],
    )

    # Set up OpenTelemetry tracing
    otlp_endpoint = settings.otel_exporter_otlp_endpoint
    if otlp_endpoint and not settings.disable_tracing:
        print(f"▶ Using OTLP tracing with endpoint: {otlp_endpoint}")
        env_name_suffix = os.getenv("ENV_NAME")
        service_name = f"letta-server-{env_name_suffix.lower()}" if env_name_suffix else "letta-server"
        from letta.otel.metrics import setup_metrics
        from letta.otel.tracing import setup_tracing

        setup_tracing(
            endpoint=otlp_endpoint,
            app=app,
            service_name=service_name,
        )
        setup_metrics(endpoint=otlp_endpoint, app=app, service_name=service_name)

        # Set up SQLAlchemy synchronous operation instrumentation
        if settings.sqlalchemy_tracing:
            from letta.otel.sqlalchemy_instrumentation_integration import setup_letta_db_instrumentation

            try:
                setup_letta_db_instrumentation(
                    enable_joined_monitoring=True,  # Monitor joined loading operations
                    sql_truncate_length=1500,  # Longer SQL statements for debugging
                )
                print("▶ SQLAlchemy synchronous operation instrumentation enabled")
            except Exception as e:
                logger.warning(f"Failed to setup SQLAlchemy instrumentation: {e}")
                # Don't fail startup if instrumentation fails

        # Ensure our validation handler overrides tracing's handler when tracing is enabled
        app.add_exception_handler(RequestValidationError, custom_request_validation_handler)

    for route in v1_routes:
        app.include_router(route, prefix=API_PREFIX)
        # this gives undocumented routes for "latest" and bare api calls.
        # we should always tie this to the newest version of the api.
        # app.include_router(route, prefix="", include_in_schema=False)
        app.include_router(route, prefix="/latest", include_in_schema=False)

    # admin/users
    app.include_router(users_router, prefix=ADMIN_PREFIX)
    app.include_router(organizations_router, prefix=ADMIN_PREFIX)

    # /api/auth endpoints
    app.include_router(setup_auth_router(server, interface, random_password), prefix=API_PREFIX)

    # / static files
    mount_static_files(app)

    no_generation = "--no-generation" in sys.argv

    # Generate OpenAPI schema after all routes are mounted
    if not no_generation:
        generate_openapi_schema(app)

    return app


app = create_application()


def start_server(
    port: Optional[int] = None,
    host: Optional[str] = None,
    debug: bool = False,
    reload: bool = False,
):
    """Convenience method to start the server from within Python"""
    if debug:
        from letta.server.server import logger as server_logger

        # Set the logging level
        server_logger.setLevel(logging.DEBUG)
        # Create a StreamHandler
        stream_handler = logging.StreamHandler()
        # Set the formatter (optional)
        formatter = logging.Formatter("%(asctime)s - %(name)s - %(levelname)s - %(message)s")
        stream_handler.setFormatter(formatter)
        # Add the handler to the logger
        server_logger.addHandler(stream_handler)

    # Experimental UV Loop Support
    try:
        if settings.use_uvloop:
            print("Running server asyncio loop on uvloop...")
            import asyncio

            import uvloop

            asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())
    except:
        pass

    if (os.getenv("LOCAL_HTTPS") == "true") or "--localhttps" in sys.argv:
        print(f"▶ Server running at: https://{host or 'localhost'}:{port or REST_DEFAULT_PORT}")
        print("▶ View using ADE at: https://app.letta.com/development-servers/local/dashboard\n")
        if importlib.util.find_spec("granian") is not None and settings.use_granian:
            from granian import Granian

            # Experimental Granian engine
            Granian(
                target="letta.server.rest_api.app:app",
                # factory=True,
                interface="asgi",
                address=host or "127.0.0.1",  # Note granian address must be an ip address
                port=port or REST_DEFAULT_PORT,
                workers=settings.uvicorn_workers,
                # runtime_blocking_threads=
                # runtime_threads=
                reload=reload or settings.uvicorn_reload,
                reload_paths=["letta/"],
                reload_ignore_worker_failure=True,
                reload_tick=4000,  # set to 4s to prevent crashing on weird state
                # log_level="info"
                ssl_keyfile="certs/localhost-key.pem",
                ssl_cert="certs/localhost.pem",
            ).serve()
        else:
            uvicorn.run(
                "letta.server.rest_api.app:app",
                host=host or "localhost",
                port=port or REST_DEFAULT_PORT,
                workers=settings.uvicorn_workers,
                reload=reload or settings.uvicorn_reload,
                timeout_keep_alive=settings.uvicorn_timeout_keep_alive,
                ssl_keyfile="certs/localhost-key.pem",
                ssl_certfile="certs/localhost.pem",
                access_log=False,
            )

    else:
        if IS_WINDOWS:
            # Windows doesn't those the fancy unicode characters
            print(f"Server running at: http://{host or 'localhost'}:{port or REST_DEFAULT_PORT}")
            print("View using ADE at: https://app.letta.com/development-servers/local/dashboard\n")
        else:
            print(f"▶ Server running at: http://{host or 'localhost'}:{port or REST_DEFAULT_PORT}")
            print("▶ View using ADE at: https://app.letta.com/development-servers/local/dashboard\n")

        if importlib.util.find_spec("granian") is not None and settings.use_granian:
            # Experimental Granian engine
            from granian import Granian

            Granian(
                target="letta.server.rest_api.app:app",
                # factory=True,
                interface="asgi",
                address=host or "127.0.0.1",  # Note granian address must be an ip address
                port=port or REST_DEFAULT_PORT,
                workers=settings.uvicorn_workers,
                # runtime_blocking_threads=
                # runtime_threads=
                reload=reload or settings.uvicorn_reload,
                reload_paths=["letta/"],
                reload_ignore_worker_failure=True,
                reload_tick=4000,  # set to 4s to prevent crashing on weird state
                # log_level="info"
            ).serve()
        else:
            uvicorn.run(
                "letta.server.rest_api.app:app",
                host=host or "localhost",
                port=port or REST_DEFAULT_PORT,
                workers=settings.uvicorn_workers,
                reload=reload or settings.uvicorn_reload,
                timeout_keep_alive=settings.uvicorn_timeout_keep_alive,
                access_log=False,
            )