Merge pull request #1074 from julep-ai/dev

dev -> main

agents-api/agents_api/clients/litellm.py

@@ -1,6 +1,7 @@
 from functools import wraps
 from typing import Literal
 
+import aiohttp
 from beartype import beartype
 from litellm import acompletion as _acompletion
 from litellm import aembedding as _aembedding
@@ -109,3 +110,26 @@ async def aembedding(
         for item in embedding_list
         if len(item["embedding"]) >= dimensions
     ]
+
+
+@beartype
+async def get_model_list(*, custom_api_key: str | None = None) -> list[dict]:
+    """
+    Fetches the list of available models from the LiteLLM server.
+
+    Returns:
+        list[dict]: A list of model information dictionaries
+    """
+
+    headers = {
+        "accept": "application/json",
+        "x-api-key": custom_api_key or litellm_master_key
+    }
+
+    async with aiohttp.ClientSession() as session, session.get(
+        url=f"{litellm_url}/models" if not custom_api_key else "/models",
+        headers=headers
+    ) as response:
+        response.raise_for_status()
+        data = await response.json()
+        return data["data"]

agents-api/agents_api/queries/chat/prepare_chat_context.py

@@ -51,7 +51,7 @@
             agents.metadata,
             agents.default_settings
         FROM session_lookup
-        INNER JOIN agents ON session_lookup.participant_id = agents.agent_id
+        INNER JOIN agents ON session_lookup.participant_id = agents.agent_id AND agents.developer_id = session_lookup.developer_id
         WHERE
             session_lookup.developer_id = $1 AND
             session_id = $2 AND
@@ -95,7 +95,7 @@
             tools.updated_at,
             tools.created_at
         FROM session_lookup
-        INNER JOIN tools ON session_lookup.participant_id = tools.agent_id
+        INNER JOIN tools ON session_lookup.participant_id = tools.agent_id AND tools.developer_id = session_lookup.developer_id
         WHERE
             session_lookup.developer_id = $1 AND
             session_id = $2 AND

agents-api/agents_api/routers/agents/create_agent.py

@@ -10,6 +10,7 @@
 )
 from ...dependencies.developer_id import get_developer_id
 from ...queries.agents.create_agent import create_agent as create_agent_query
+from ..utils.model_validation import validate_model
 from .router import router
 
 
@@ -18,7 +19,10 @@ async def create_agent(
     x_developer_id: Annotated[UUID, Depends(get_developer_id)],
     data: CreateAgentRequest,
 ) -> ResourceCreatedResponse:
-    # TODO: Validate model name
+
+    if data.model:
+        await validate_model(data.model)
+
     agent = await create_agent_query(
         developer_id=x_developer_id,
         data=data,

agents-api/agents_api/routers/agents/create_or_update_agent.py

@@ -12,6 +12,7 @@
 from ...queries.agents.create_or_update_agent import (
     create_or_update_agent as create_or_update_agent_query,
 )
+from ..utils.model_validation import validate_model
 from .router import router
 
 
@@ -21,7 +22,10 @@ async def create_or_update_agent(
     data: CreateOrUpdateAgentRequest,
     x_developer_id: Annotated[UUID, Depends(get_developer_id)],
 ) -> ResourceCreatedResponse:
-    # TODO: Validate model name
+
+    if data.model:
+        await validate_model(data.model)
+
     agent = await create_or_update_agent_query(
         developer_id=x_developer_id,
         agent_id=agent_id,

agents-api/agents_api/routers/agents/patch_agent.py

@@ -7,6 +7,7 @@
 from ...autogen.openapi_model import PatchAgentRequest, ResourceUpdatedResponse
 from ...dependencies.developer_id import get_developer_id
 from ...queries.agents.patch_agent import patch_agent as patch_agent_query
+from ..utils.model_validation import validate_model
 from .router import router
 
 
@@ -21,6 +22,10 @@ async def patch_agent(
     agent_id: UUID,
     data: PatchAgentRequest,
 ) -> ResourceUpdatedResponse:
+
+    if data.model:
+        await validate_model(data.model)
+
     return await patch_agent_query(
         agent_id=agent_id,
         developer_id=x_developer_id,

agents-api/agents_api/routers/agents/update_agent.py

@@ -7,6 +7,7 @@
 from ...autogen.openapi_model import ResourceUpdatedResponse, UpdateAgentRequest
 from ...dependencies.developer_id import get_developer_id
 from ...queries.agents.update_agent import update_agent as update_agent_query
+from ..utils.model_validation import validate_model
 from .router import router
 
 
@@ -20,7 +21,11 @@ async def update_agent(
     x_developer_id: Annotated[UUID, Depends(get_developer_id)],
     agent_id: UUID,
     data: UpdateAgentRequest,
-) -> ResourceUpdatedResponse:
+    ) -> ResourceUpdatedResponse:
+
+    if data.model:
+        await validate_model(data.model)
+
     return await update_agent_query(
         developer_id=x_developer_id,
         agent_id=agent_id,

agents-api/agents_api/routers/sessions/chat.py

@@ -23,6 +23,7 @@
 from ...queries.chat.prepare_chat_context import prepare_chat_context
 from ...queries.entries.create_entries import create_entries
 from ...queries.sessions.count_sessions import count_sessions as count_sessions_query
+from ..utils.model_validation import validate_model
 from .metrics import total_tokens_per_user
 from .router import router
 
@@ -55,6 +56,10 @@ async def chat(
     Returns:
         ChatResponse: The chat response.
     """
+
+    if chat_input.model:
+        await validate_model(chat_input.model)
+
     # check if the developer is paid
     if "paid" not in developer.tags:
         # get the session length

agents-api/agents_api/routers/utils/model_validation.py

@@ -0,0 +1,19 @@
+from fastapi import HTTPException
+from starlette.status import HTTP_400_BAD_REQUEST
+
+from ...clients.litellm import get_model_list
+
+
+async def validate_model(model_name: str) -> None:
+    """
+    Validates if a given model name is available in LiteLLM.
+    Raises HTTPException if model is not available.
+    """
+    models = await get_model_list()
+    available_models = [model["id"] for model in models]
+
+    if model_name not in available_models:
+        raise HTTPException(
+            status_code=HTTP_400_BAD_REQUEST,
+            detail=f"Model {model_name} not available. Available models: {available_models}"
+        )

agents-api/tests/fixtures.py

@@ -2,6 +2,7 @@
 import random
 import string
 import sys
+from unittest.mock import patch
 from uuid import UUID
 
 from agents_api.autogen.openapi_model import (
@@ -440,10 +441,18 @@ async def test_tool(
     return tool
 
 
+SAMPLE_MODELS = [
+    {"id": "gpt-4"},
+    {"id": "gpt-3.5-turbo"},
+    {"id": "gpt-4o-mini"},
+]
+
+
 @fixture(scope="global")
 def client(_dsn=pg_dsn):
     with TestClient(app=app) as client:
-        yield client
+        with patch("agents_api.routers.utils.model_validation.get_model_list", return_value=SAMPLE_MODELS):
+            yield client
 
 
 @fixture(scope="global")

agents-api/tests/test_model_validation.py

@@ -0,0 +1,28 @@
+from unittest.mock import patch
+
+from agents_api.routers.utils.model_validation import validate_model
+from fastapi import HTTPException
+from ward import raises, test
+
+from tests.fixtures import SAMPLE_MODELS
+
+
+@test("validate_model: succeeds when model is available")
+async def _():
+    # Use async context manager for patching
+    with patch("agents_api.routers.utils.model_validation.get_model_list") as mock_get_models:
+        mock_get_models.return_value = SAMPLE_MODELS
+        await validate_model("gpt-4o-mini")
+        mock_get_models.assert_called_once()
+
+
+@test("validate_model: fails when model is unavailable")
+async def _():
+    with patch("agents_api.routers.utils.model_validation.get_model_list") as mock_get_models:
+        mock_get_models.return_value = SAMPLE_MODELS
+        with raises(HTTPException) as exc:
+            await validate_model("non-existent-model")
+
+        assert exc.raised.status_code == 400
+        assert "Model non-existent-model not available" in exc.raised.detail
+        mock_get_models.assert_called_once()