letta-nightly 0.4.1.dev20241008104105__py3-none-any.whl → 0.4.1.dev20241010104112__py3-none-any.whl

letta/llm_api/llm_api_tools.py

@@ -28,7 +28,6 @@ from letta.local_llm.constants import (
     INNER_THOUGHTS_KWARG,
     INNER_THOUGHTS_KWARG_DESCRIPTION,
 )
-from letta.providers import GoogleAIProvider
 from letta.schemas.enums import OptionState
 from letta.schemas.llm_config import LLMConfig
 from letta.schemas.message import Message
@@ -189,6 +188,9 @@ def create(
         if model_settings.azure_base_url is None:
             raise ValueError(f"Azure base url is missing. Did you set AZURE_BASE_URL in your env?")
 
+        if model_settings.azure_api_version is None:
+            raise ValueError(f"Azure API version is missing. Did you set AZURE_API_VERSION in your env?")
+
         # Set the llm config model_endpoint from model_settings
         # For Azure, this model_endpoint is required to be configured via env variable, so users don't need to provide it in the LLM config
         llm_config.model_endpoint = model_settings.azure_base_url
@@ -215,20 +217,15 @@ def create(
         if not use_tool_naming:
             raise NotImplementedError("Only tool calling supported on Google AI API requests")
 
-        # NOTE: until Google AI supports CoT / text alongside function calls,
-        # we need to put it in a kwarg (unless we want to split the message into two)
-        google_ai_inner_thoughts_in_kwarg = True
-
         if functions is not None:
             tools = [{"type": "function", "function": f} for f in functions]
             tools = [Tool(**t) for t in tools]
-            tools = convert_tools_to_google_ai_format(tools, inner_thoughts_in_kwargs=google_ai_inner_thoughts_in_kwarg)
+            tools = convert_tools_to_google_ai_format(tools, inner_thoughts_in_kwargs=True)
         else:
             tools = None
 
         return google_ai_chat_completions_request(
-            inner_thoughts_in_kwargs=google_ai_inner_thoughts_in_kwarg,
-            service_endpoint=GoogleAIProvider(model_settings.gemini_api_key).service_endpoint,
+            base_url=llm_config.model_endpoint,
             model=llm_config.model,
             api_key=model_settings.gemini_api_key,
             # see structure of payload here: https://ai.google.dev/docs/function_calling
@@ -236,6 +233,7 @@ def create(
                 contents=[m.to_google_ai_dict() for m in messages],
                 tools=tools,
             ),
+            inner_thoughts_in_kwargs=True,
         )
 
     elif llm_config.model_endpoint_type == "anthropic":
@@ -244,12 +242,6 @@ def create(
         if not use_tool_naming:
             raise NotImplementedError("Only tool calling supported on Anthropic API requests")
 
-        if functions is not None:
-            tools = [{"type": "function", "function": f} for f in functions]
-            tools = [Tool(**t) for t in tools]
-        else:
-            tools = None
-
         return anthropic_chat_completions_request(
             url=llm_config.model_endpoint,
             api_key=model_settings.anthropic_api_key,
@@ -296,7 +288,6 @@ def create(
             raise NotImplementedError(f"Streaming not yet implemented for Groq.")
 
         if model_settings.groq_api_key is None and llm_config.model_endpoint == "https://api.groq.com/openai/v1/chat/completions":
-            # only is a problem if we are *not* using an openai proxy
             raise ValueError(f"Groq key is missing from letta config file")
 
         # force to true for groq, since they don't support 'content' is non-null

letta/llm_api/openai.py

@@ -9,7 +9,7 @@ from httpx_sse._exceptions import SSEError
 
 from letta.constants import OPENAI_CONTEXT_WINDOW_ERROR_SUBSTRING
 from letta.errors import LLMError
-from letta.llm_api.helpers import add_inner_thoughts_to_functions
+from letta.llm_api.helpers import add_inner_thoughts_to_functions, make_post_request
 from letta.local_llm.constants import (
     INNER_THOUGHTS_KWARG,
     INNER_THOUGHTS_KWARG_DESCRIPTION,
@@ -145,6 +145,7 @@ def build_openai_chat_completions_request(
         import uuid
 
         data.user = str(uuid.UUID(int=0))
+        data.model = "memgpt-openai"
 
     return data
 
@@ -483,58 +484,14 @@ def openai_chat_completions_request(
         data.pop("tools")
         data.pop("tool_choice", None)  # extra safe,  should exist always (default="auto")
 
-    printd(f"Sending request to {url}")
-    try:
-        response = requests.post(url, headers=headers, json=data)
-        printd(f"response = {response}, response.text = {response.text}")
-        # print(json.dumps(data, indent=4))
-        # raise requests.exceptions.HTTPError
-        response.raise_for_status()  # Raises HTTPError for 4XX/5XX status
-
-        response = response.json()  # convert to dict from string
-        printd(f"response.json = {response}")
-
-        response = ChatCompletionResponse(**response)  # convert to 'dot-dict' style which is the openai python client default
-        return response
-    except requests.exceptions.HTTPError as http_err:
-        # Handle HTTP errors (e.g., response 4XX, 5XX)
-        printd(f"Got HTTPError, exception={http_err}, payload={data}")
-        raise http_err
-    except requests.exceptions.RequestException as req_err:
-        # Handle other requests-related errors (e.g., connection error)
-        printd(f"Got RequestException, exception={req_err}")
-        raise req_err
-    except Exception as e:
-        # Handle other potential errors
-        printd(f"Got unknown Exception, exception={e}")
-        raise e
+    response_json = make_post_request(url, headers, data)
+    return ChatCompletionResponse(**response_json)
 
 
 def openai_embeddings_request(url: str, api_key: str, data: dict) -> EmbeddingResponse:
     """https://platform.openai.com/docs/api-reference/embeddings/create"""
-    from letta.utils import printd
 
     url = smart_urljoin(url, "embeddings")
     headers = {"Content-Type": "application/json", "Authorization": f"Bearer {api_key}"}
-
-    printd(f"Sending request to {url}")
-    try:
-        response = requests.post(url, headers=headers, json=data)
-        printd(f"response = {response}")
-        response.raise_for_status()  # Raises HTTPError for 4XX/5XX status
-        response = response.json()  # convert to dict from string
-        printd(f"response.json = {response}")
-        response = EmbeddingResponse(**response)  # convert to 'dot-dict' style which is the openai python client default
-        return response
-    except requests.exceptions.HTTPError as http_err:
-        # Handle HTTP errors (e.g., response 4XX, 5XX)
-        printd(f"Got HTTPError, exception={http_err}, payload={data}")
-        raise http_err
-    except requests.exceptions.RequestException as req_err:
-        # Handle other requests-related errors (e.g., connection error)
-        printd(f"Got RequestException, exception={req_err}")
-        raise req_err
-    except Exception as e:
-        # Handle other potential errors
-        printd(f"Got unknown Exception, exception={e}")
-        raise e
+    response_json = make_post_request(url, headers, data)
+    return EmbeddingResponse(**response_json)

letta/local_llm/constants.py

@@ -29,3 +29,6 @@ DEFAULT_WRAPPER_NAME = "chatml"
 
 INNER_THOUGHTS_KWARG = "inner_thoughts"
 INNER_THOUGHTS_KWARG_DESCRIPTION = "Deep inner monologue private to you only."
+INNER_THOUGHTS_CLI_SYMBOL = "💭"
+
+ASSISTANT_MESSAGE_CLI_SYMBOL = "🤖"

letta/main.py

@@ -366,7 +366,7 @@ def run_agent_loop(
                 first_message=False,
                 skip_verify=no_verify,
                 stream=stream,
-                inner_thoughts_in_kwargs=inner_thoughts_in_kwargs,
+                inner_thoughts_in_kwargs_option=inner_thoughts_in_kwargs,
                 ms=ms,
             )
             new_messages = step_response.messages

letta/metadata.py

@@ -218,6 +218,7 @@ class AgentModel(Base):
     tools = Column(JSON)
 
     # configs
+    agent_type = Column(String)
     llm_config = Column(LLMConfigColumn)
     embedding_config = Column(EmbeddingConfigColumn)
 
@@ -243,6 +244,7 @@ class AgentModel(Base):
             memory=Memory.load(self.memory),  # load dictionary
             system=self.system,
             tools=self.tools,
+            agent_type=self.agent_type,
             llm_config=self.llm_config,
             embedding_config=self.embedding_config,
             metadata_=self.metadata_,

letta/providers.py

@@ -1,14 +1,18 @@
 from typing import List, Optional
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, model_validator
 
 from letta.constants import LLM_MAX_TOKENS
+from letta.llm_api.azure_openai import (
+    get_azure_chat_completions_endpoint,
+    get_azure_embeddings_endpoint,
+)
+from letta.llm_api.azure_openai_constants import AZURE_MODEL_TO_CONTEXT_LENGTH
 from letta.schemas.embedding_config import EmbeddingConfig
 from letta.schemas.llm_config import LLMConfig
 
 
 class Provider(BaseModel):
-    base_url: str
 
     def list_llm_models(self):
         return []
@@ -20,6 +24,32 @@ class Provider(BaseModel):
         pass
 
 
+class LettaProvider(Provider):
+
+    name: str = "letta"
+
+    def list_llm_models(self) -> List[LLMConfig]:
+        return [
+            LLMConfig(
+                model="letta-free",  # NOTE: renamed
+                model_endpoint_type="openai",
+                model_endpoint="https://inference.memgpt.ai",
+                context_window=16384,
+            )
+        ]
+
+    def list_embedding_models(self):
+        return [
+            EmbeddingConfig(
+                embedding_model="letta-free",  # NOTE: renamed
+                embedding_endpoint_type="hugging-face",
+                embedding_endpoint="https://embeddings.memgpt.ai",
+                embedding_dim=1024,
+                embedding_chunk_size=300,
+            )
+        ]
+
+
 class OpenAIProvider(Provider):
     name: str = "openai"
     api_key: str = Field(..., description="API key for the OpenAI API.")
@@ -122,34 +152,64 @@ class OllamaProvider(OpenAIProvider):
         response = requests.post(f"{self.base_url}/api/show", json={"name": model_name, "verbose": True})
         response_json = response.json()
 
-        # thank you vLLM: https://github.com/vllm-project/vllm/blob/main/vllm/config.py#L1675
-        possible_keys = [
-            # OPT
-            "max_position_embeddings",
-            # GPT-2
-            "n_positions",
-            # MPT
-            "max_seq_len",
-            # ChatGLM2
-            "seq_length",
-            # Command-R
-            "model_max_length",
-            # Others
-            "max_sequence_length",
-            "max_seq_length",
-            "seq_len",
-        ]
-
+        ## thank you vLLM: https://github.com/vllm-project/vllm/blob/main/vllm/config.py#L1675
+        # possible_keys = [
+        #    # OPT
+        #    "max_position_embeddings",
+        #    # GPT-2
+        #    "n_positions",
+        #    # MPT
+        #    "max_seq_len",
+        #    # ChatGLM2
+        #    "seq_length",
+        #    # Command-R
+        #    "model_max_length",
+        #    # Others
+        #    "max_sequence_length",
+        #    "max_seq_length",
+        #    "seq_len",
+        # ]
         # max_position_embeddings
         # parse model cards: nous, dolphon, llama
         for key, value in response_json["model_info"].items():
-            if "context_window" in key:
+            if "context_length" in key:
+                return value
+        return None
+
+    def get_model_embedding_dim(self, model_name: str):
+        import requests
+
+        response = requests.post(f"{self.base_url}/api/show", json={"name": model_name, "verbose": True})
+        response_json = response.json()
+        for key, value in response_json["model_info"].items():
+            if "embedding_length" in key:
                 return value
         return None
 
     def list_embedding_models(self) -> List[EmbeddingConfig]:
-        # TODO: filter embedding models
-        return []
+        # https://github.com/ollama/ollama/blob/main/docs/api.md#list-local-models
+        import requests
+
+        response = requests.get(f"{self.base_url}/api/tags")
+        if response.status_code != 200:
+            raise Exception(f"Failed to list Ollama models: {response.text}")
+        response_json = response.json()
+
+        configs = []
+        for model in response_json["models"]:
+            embedding_dim = self.get_model_embedding_dim(model["name"])
+            if not embedding_dim:
+                continue
+            configs.append(
+                EmbeddingConfig(
+                    embedding_model=model["name"],
+                    embedding_endpoint_type="ollama",
+                    embedding_endpoint=self.base_url,
+                    embedding_dim=embedding_dim,
+                    embedding_chunk_size=300,
+                )
+            )
+        return configs
 
 
 class GroqProvider(OpenAIProvider):
@@ -182,20 +242,21 @@ class GroqProvider(OpenAIProvider):
 class GoogleAIProvider(Provider):
     # gemini
     api_key: str = Field(..., description="API key for the Google AI API.")
-    service_endpoint: str = "generativelanguage"
     base_url: str = "https://generativelanguage.googleapis.com"
 
     def list_llm_models(self):
         from letta.llm_api.google_ai import google_ai_get_model_list
 
-        # TODO: use base_url instead
-        model_options = google_ai_get_model_list(service_endpoint=self.service_endpoint, api_key=self.api_key)
+        model_options = google_ai_get_model_list(base_url=self.base_url, api_key=self.api_key)
+        # filter by 'generateContent' models
+        model_options = [mo for mo in model_options if "generateContent" in mo["supportedGenerationMethods"]]
         model_options = [str(m["name"]) for m in model_options]
+
+        # filter by model names
         model_options = [mo[len("models/") :] if mo.startswith("models/") else mo for mo in model_options]
+
         # TODO remove manual filtering for gemini-pro
         model_options = [mo for mo in model_options if str(mo).startswith("gemini") and "-pro" in str(mo)]
-        # TODO: add context windows
-        # model_options = ["gemini-pro"]
 
         configs = []
         for model in model_options:
@@ -210,21 +271,94 @@ class GoogleAIProvider(Provider):
         return configs
 
     def list_embedding_models(self):
-        return []
+        from letta.llm_api.google_ai import google_ai_get_model_list
+
+        # TODO: use base_url instead
+        model_options = google_ai_get_model_list(base_url=self.base_url, api_key=self.api_key)
+        # filter by 'generateContent' models
+        model_options = [mo for mo in model_options if "embedContent" in mo["supportedGenerationMethods"]]
+        model_options = [str(m["name"]) for m in model_options]
+        model_options = [mo[len("models/") :] if mo.startswith("models/") else mo for mo in model_options]
+
+        configs = []
+        for model in model_options:
+            configs.append(
+                EmbeddingConfig(
+                    embedding_model=model,
+                    embedding_endpoint_type="google_ai",
+                    embedding_endpoint=self.base_url,
+                    embedding_dim=768,
+                    embedding_chunk_size=300,  # NOTE: max is 2048
+                )
+            )
+        return configs
 
     def get_model_context_window(self, model_name: str):
         from letta.llm_api.google_ai import google_ai_get_model_context_window
 
-        # TODO: use base_url instead
-        return google_ai_get_model_context_window(self.service_endpoint, self.api_key, model_name)
+        return google_ai_get_model_context_window(self.base_url, self.api_key, model_name)
 
 
 class AzureProvider(Provider):
     name: str = "azure"
+    latest_api_version: str = "2024-09-01-preview"  # https://learn.microsoft.com/en-us/azure/ai-services/openai/api-version-deprecation
     base_url: str = Field(
         ..., description="Base URL for the Azure API endpoint. This should be specific to your org, e.g. `https://letta.openai.azure.com`."
     )
     api_key: str = Field(..., description="API key for the Azure API.")
+    api_version: str = Field(latest_api_version, description="API version for the Azure API")
+
+    @model_validator(mode="before")
+    def set_default_api_version(cls, values):
+        """
+        This ensures that api_version is always set to the default if None is passed in.
+        """
+        if values.get("api_version") is None:
+            values["api_version"] = cls.model_fields["latest_api_version"].default
+        return values
+
+    def list_llm_models(self) -> List[LLMConfig]:
+        from letta.llm_api.azure_openai import (
+            azure_openai_get_chat_completion_model_list,
+        )
+
+        model_options = azure_openai_get_chat_completion_model_list(self.base_url, api_key=self.api_key, api_version=self.api_version)
+        configs = []
+        for model_option in model_options:
+            model_name = model_option["id"]
+            context_window_size = self.get_model_context_window(model_name)
+            model_endpoint = get_azure_chat_completions_endpoint(self.base_url, model_name, self.api_version)
+            configs.append(
+                LLMConfig(model=model_name, model_endpoint_type="azure", model_endpoint=model_endpoint, context_window=context_window_size)
+            )
+        return configs
+
+    def list_embedding_models(self) -> List[EmbeddingConfig]:
+        from letta.llm_api.azure_openai import azure_openai_get_embeddings_model_list
+
+        model_options = azure_openai_get_embeddings_model_list(
+            self.base_url, api_key=self.api_key, api_version=self.api_version, require_embedding_in_name=True
+        )
+        configs = []
+        for model_option in model_options:
+            model_name = model_option["id"]
+            model_endpoint = get_azure_embeddings_endpoint(self.base_url, model_name, self.api_version)
+            configs.append(
+                EmbeddingConfig(
+                    embedding_model=model_name,
+                    embedding_endpoint_type="azure",
+                    embedding_endpoint=model_endpoint,
+                    embedding_dim=768,
+                    embedding_chunk_size=300,  # NOTE: max is 2048
+                )
+            )
+        return configs
+
+    def get_model_context_window(self, model_name: str):
+        """
+        This is hardcoded for now, since there is no API endpoints to retrieve metadata for a model.
+        """
+        return AZURE_MODEL_TO_CONTEXT_LENGTH.get(model_name, 4096)
 
 
 class VLLMProvider(OpenAIProvider):

letta/schemas/agent.py

@@ -1,5 +1,6 @@
 import uuid
 from datetime import datetime
+from enum import Enum
 from typing import Dict, List, Optional, Union
 
 from pydantic import BaseModel, Field, field_validator
@@ -21,6 +22,15 @@ class BaseAgent(LettaBase, validate_assignment=True):
     user_id: Optional[str] = Field(None, description="The user id of the agent.")
 
 
+class AgentType(str, Enum):
+    """
+    Enum to represent the type of agent.
+    """
+
+    memgpt_agent = "memgpt_agent"
+    split_thread_agent = "split_thread_agent"
+
+
 class AgentState(BaseAgent):
     """
     Representation of an agent's state. This is the state of the agent at a given time, and is persisted in the DB backend. The state has all the information needed to recreate a persisted agent.
@@ -52,6 +62,9 @@ class AgentState(BaseAgent):
     # system prompt
     system: str = Field(..., description="The system prompt used by the agent.")
 
+    # agent configuration
+    agent_type: AgentType = Field(..., description="The type of agent.")
+
     # llm information
     llm_config: LLMConfig = Field(..., description="The LLM configuration used by the agent.")
     embedding_config: EmbeddingConfig = Field(..., description="The embedding configuration used by the agent.")
@@ -64,6 +77,7 @@ class CreateAgent(BaseAgent):
     memory: Optional[Memory] = Field(None, description="The in-context memory of the agent.")
     tools: Optional[List[str]] = Field(None, description="The tools used by the agent.")
     system: Optional[str] = Field(None, description="The system prompt used by the agent.")
+    agent_type: Optional[AgentType] = Field(None, description="The type of agent.")
     llm_config: Optional[LLMConfig] = Field(None, description="The LLM configuration used by the agent.")
     embedding_config: Optional[EmbeddingConfig] = Field(None, description="The embedding configuration used by the agent.")
 

letta/schemas/llm_config.py

@@ -35,9 +35,6 @@ class LLMConfig(BaseModel):
         "hugging-face",
     ] = Field(..., description="The endpoint type for the model.")
     model_endpoint: Optional[str] = Field(None, description="The endpoint for the model.")
-    api_version: Optional[str] = Field(
-        None, description="The version for the model API. Used by the Azure provider backend, e.g. 2023-03-15-preview."
-    )
     model_wrapper: Optional[str] = Field(None, description="The wrapper for the model.")
     context_window: int = Field(..., description="The context window size for the model.")
 

letta/schemas/openai/chat_completion_response.py

@@ -74,6 +74,9 @@ class ChatCompletionResponse(BaseModel):
     object: Literal["chat.completion"] = "chat.completion"
     usage: UsageStatistics
 
+    def __str__(self):
+        return self.model_dump_json(indent=4)
+
 
 class FunctionCallDelta(BaseModel):
     # arguments: Optional[str] = None

letta/schemas/tool.py

@@ -93,7 +93,7 @@ class Tool(BaseTool):
         # append heartbeat (necessary for triggering another reasoning step after this tool call)
         json_schema["parameters"]["properties"]["request_heartbeat"] = {
             "type": "boolean",
-            "description": "Request an immediate heartbeat after function execution. Set to 'true' if you want to send a follow-up message or run a follow-up function.",
+            "description": "Request an immediate heartbeat after function execution. Set to `True` if you want to send a follow-up message or run a follow-up function.",
         }
         json_schema["parameters"]["required"].append("request_heartbeat")
 
@@ -128,7 +128,7 @@ class Tool(BaseTool):
         # append heartbeat (necessary for triggering another reasoning step after this tool call)
         json_schema["parameters"]["properties"]["request_heartbeat"] = {
             "type": "boolean",
-            "description": "Request an immediate heartbeat after function execution. Set to 'true' if you want to send a follow-up message or run a follow-up function.",
+            "description": "Request an immediate heartbeat after function execution. Set to `True` if you want to send a follow-up message or run a follow-up function.",
         }
         json_schema["parameters"]["required"].append("request_heartbeat")
 
@@ -161,7 +161,7 @@ class Tool(BaseTool):
         # append heartbeat (necessary for triggering another reasoning step after this tool call)
         json_schema["parameters"]["properties"]["request_heartbeat"] = {
             "type": "boolean",
-            "description": "Request an immediate heartbeat after function execution. Set to 'true' if you want to send a follow-up message or run a follow-up function.",
+            "description": "Request an immediate heartbeat after function execution. Set to `True` if you want to send a follow-up message or run a follow-up function.",
         }
         json_schema["parameters"]["required"].append("request_heartbeat")
 

letta/server/rest_api/routers/openai/assistants/threads.py

@@ -1,5 +1,5 @@
 import uuid
-from typing import TYPE_CHECKING, List
+from typing import TYPE_CHECKING, List, Optional
 
 from fastapi import APIRouter, Body, Depends, Header, HTTPException, Path, Query
 
@@ -43,7 +43,7 @@ router = APIRouter(prefix="/v1/threads", tags=["threads"])
 def create_thread(
     request: CreateThreadRequest = Body(...),
     server: SyncServer = Depends(get_letta_server),
-    user_id: str = Header(None),  # Extract user_id from header, default to None if not present
+    user_id: Optional[str] = Header(None, alias="user_id"),  # Extract user_id from header, default to None if not present
 ):
     # TODO: use requests.description and requests.metadata fields
     # TODO: handle requests.file_ids and requests.tools
@@ -68,7 +68,7 @@ def create_thread(
 def retrieve_thread(
     thread_id: str = Path(..., description="The unique identifier of the thread."),
     server: SyncServer = Depends(get_letta_server),
-    user_id: str = Header(None),  # Extract user_id from header, default to None if not present
+    user_id: Optional[str] = Header(None, alias="user_id"),  # Extract user_id from header, default to None if not present
 ):
     actor = server.get_user_or_default(user_id=user_id)
     agent = server.get_agent(user_id=actor.id, agent_id=thread_id)
@@ -102,7 +102,7 @@ def create_message(
     thread_id: str = Path(..., description="The unique identifier of the thread."),
     request: CreateMessageRequest = Body(...),
     server: SyncServer = Depends(get_letta_server),
-    user_id: str = Header(None),  # Extract user_id from header, default to None if not present
+    user_id: Optional[str] = Header(None, alias="user_id"),  # Extract user_id from header, default to None if not present
 ):
     actor = server.get_user_or_default(user_id=user_id)
     agent_id = thread_id
@@ -146,7 +146,7 @@ def list_messages(
     after: str = Query(None, description="A cursor for use in pagination. `after` is an object ID that defines your place in the list."),
     before: str = Query(None, description="A cursor for use in pagination. `after` is an object ID that defines your place in the list."),
     server: SyncServer = Depends(get_letta_server),
-    user_id: str = Header(None),  # Extract user_id from header, default to None if not present
+    user_id: Optional[str] = Header(None, alias="user_id"),  # Extract user_id from header, default to None if not present
 ):
     actor = server.get_user_or_default(user_id)
     after_uuid = after if before else None

letta/server/rest_api/routers/openai/chat_completions/chat_completions.py

@@ -1,5 +1,5 @@
 import json
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Optional
 
 from fastapi import APIRouter, Body, Depends, Header, HTTPException
 
@@ -30,7 +30,7 @@ router = APIRouter(prefix="/v1/chat/completions", tags=["chat_completions"])
 async def create_chat_completion(
     completion_request: ChatCompletionRequest = Body(...),
     server: "SyncServer" = Depends(get_letta_server),
-    user_id: str = Header(None),  # Extract user_id from header, default to None if not present
+    user_id: Optional[str] = Header(None, alias="user_id"),  # Extract user_id from header, default to None if not present
 ):
     """Send a message to a Letta agent via a /chat/completions completion_request
     The bearer token will be used to identify the user.