letta-nightly 0.5.1.dev20241029104141__tar.gz → 0.5.1.dev20241030104135__tar.gz

{letta_nightly-0.5.1.dev20241029104141 → letta_nightly-0.5.1.dev20241030104135}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: letta-nightly
-Version: 0.5.1.dev20241029104141
+Version: 0.5.1.dev20241030104135
 Summary: Create LLM agents with long-term memory and custom tools
 License: Apache License
 Author: Letta Team

{letta_nightly-0.5.1.dev20241029104141 → letta_nightly-0.5.1.dev20241030104135}/letta/server/rest_api/routers/v1/agents.py

@@ -359,7 +359,7 @@ def update_message(
     return server.update_agent_message(agent_id=agent_id, request=request)
 
 
-@router.post("/{agent_id}/messages", response_model=LettaResponse, operation_id="create_agent_message")
+@router.post("/{agent_id}/messages", response_model=None, operation_id="create_agent_message")
 async def send_message(
     agent_id: str,
     server: SyncServer = Depends(get_letta_server),
@@ -373,16 +373,10 @@ async def send_message(
     """
     actor = server.get_user_or_default(user_id=user_id)
 
-    # TODO(charles): support sending multiple messages
-    assert len(request.messages) == 1, f"Multiple messages not supported: {request.messages}"
-    request.messages[0]
-
     return await send_message_to_agent(
         server=server,
         agent_id=agent_id,
         user_id=actor.id,
-        # role=message.role,
-        # message=message.text,
         messages=request.messages,
         stream_steps=request.stream_steps,
         stream_tokens=request.stream_tokens,
@@ -463,8 +457,12 @@ async def send_message_to_agent(
         # Offload the synchronous message_func to a separate thread
         streaming_interface.stream_start()
         task = asyncio.create_task(
-            # asyncio.to_thread(message_func, user_id=user_id, agent_id=agent_id, message=message, timestamp=timestamp)
-            asyncio.to_thread(server.send_messages, user_id=user_id, agent_id=agent_id, messages=messages)
+            asyncio.to_thread(
+                server.send_messages,
+                user_id=user_id,
+                agent_id=agent_id,
+                messages=messages,
+            )
         )
 
         if stream_steps:
@@ -474,7 +472,11 @@ async def send_message_to_agent(
 
             # return a stream
             return StreamingResponse(
-                sse_async_generator(streaming_interface.get_generator(), finish_message=include_final_message),
+                sse_async_generator(
+                    streaming_interface.get_generator(),
+                    usage_task=task,
+                    finish_message=include_final_message,
+                ),
                 media_type="text/event-stream",
             )
 

{letta_nightly-0.5.1.dev20241029104141 → letta_nightly-0.5.1.dev20241030104135}/letta/server/rest_api/utils.py

@@ -1,10 +1,13 @@
+import asyncio
 import json
 import traceback
+import warnings
 from enum import Enum
-from typing import AsyncGenerator, Union
+from typing import AsyncGenerator, Optional, Union
 
 from pydantic import BaseModel
 
+from letta.schemas.usage import LettaUsageStatistics
 from letta.server.rest_api.interface import StreamingServerInterface
 from letta.server.server import SyncServer
 
@@ -24,7 +27,11 @@ def sse_formatter(data: Union[dict, str]) -> str:
     return f"data: {data_str}\n\n"
 
 
-async def sse_async_generator(generator: AsyncGenerator, finish_message=True):
+async def sse_async_generator(
+    generator: AsyncGenerator,
+    usage_task: Optional[asyncio.Task] = None,
+    finish_message=True,
+):
     """
     Wraps a generator for use in Server-Sent Events (SSE), handling errors and ensuring a completion message.
 
@@ -45,6 +52,18 @@ async def sse_async_generator(generator: AsyncGenerator, finish_message=True):
                 chunk = str(chunk)
             yield sse_formatter(chunk)
 
+        # If we have a usage task, wait for it and send its result
+        if usage_task is not None:
+            try:
+                usage = await usage_task
+                # Double-check the type
+                if not isinstance(usage, LettaUsageStatistics):
+                    raise ValueError(f"Expected LettaUsageStatistics, got {type(usage)}")
+                yield sse_formatter({"usage": usage.model_dump()})
+            except Exception as e:
+                warnings.warn(f"Error getting usage data: {e}")
+                yield sse_formatter({"error": "Failed to get usage data"})
+
     except Exception as e:
         print("stream decoder hit error:", e)
         print(traceback.print_stack())

{letta_nightly-0.5.1.dev20241029104141 → letta_nightly-0.5.1.dev20241030104135}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "letta-nightly"
-version = "0.5.1.dev20241029104141"
+version = "0.5.1.dev20241030104135"
 packages = [
     {include = "letta"}
 ]