fix: upgrade google-genai sdk version and fix gemini 3 streaming (#6437)

* base

* base

---------

Co-authored-by: Letta Bot <noreply@letta.com>

letta/adapters/simple_llm_stream_adapter.py

@@ -112,6 +112,11 @@ class SimpleLLMStreamAdapter(LettaLLMStreamAdapter):
 
         # Start the streaming request (map provider errors to common LLMError types)
         try:
+            # Gemini uses async generator pattern (no await) to maintain connection lifecycle
+            # Other providers return awaitables that resolve to iterators
+            if self.llm_config.model_endpoint_type in [ProviderType.google_ai, ProviderType.google_vertex]:
+                stream = self.llm_client.stream_async(request_data, self.llm_config)
+            else:
                 stream = await self.llm_client.stream_async(request_data, self.llm_config)
         except Exception as e:
             raise self.llm_client.handle_llm_error(e)

letta/llm_api/google_vertex_client.py

@@ -3,8 +3,7 @@ import json
 import uuid
 from typing import AsyncIterator, List, Optional
 
-from google import genai
-from google.genai import errors
+from google.genai import Client, errors
 from google.genai.types import (
     FunctionCallingConfig,
     FunctionCallingConfigMode,
@@ -51,7 +50,7 @@ class GoogleVertexClient(LLMClientBase):
 
     def _get_client(self):
         timeout_ms = int(settings.llm_request_timeout_seconds * 1000)
-        return genai.Client(
+        return Client(
             vertexai=True,
             project=model_settings.google_cloud_project,
             location=model_settings.google_cloud_location,
@@ -142,11 +141,15 @@ class GoogleVertexClient(LLMClientBase):
     @trace_method
     async def stream_async(self, request_data: dict, llm_config: LLMConfig) -> AsyncIterator[GenerateContentResponse]:
         client = self._get_client()
-        return await client.aio.models.generate_content_stream(
+        response = await client.aio.models.generate_content_stream(
             model=llm_config.model,
             contents=request_data["contents"],
             config=request_data["config"],
         )
+        # Direct yield - keeps response alive in generator's local scope throughout iteration
+        # This is required because the SDK's connection lifecycle is tied to the response object
+        async for chunk in response:
+            yield chunk
 
     @staticmethod
     def add_dummy_model_messages(messages: List[dict]) -> List[dict]:

pyproject.toml

@@ -69,7 +69,7 @@ dependencies = [
     "ruff[dev]>=0.12.10",
     "trafilatura",
     "readability-lxml",
-    "google-genai>=1.15.0",
+    "google-genai>=1.52.0",
     "datadog>=0.49.1",
     "psutil>=5.9.0",
 ]

uv.lock

@@ -1487,7 +1487,7 @@ wheels = [
 
 [[package]]
 name = "google-genai"
-version = "1.31.0"
+version = "1.52.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "anyio" },
@@ -1499,9 +1499,9 @@ dependencies = [
     { name = "typing-extensions" },
     { name = "websockets" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/e0/1b/da30fa6e2966942d7028a58eb7aa7d04544dcc3aa66194365b2e0adac570/google_genai-1.31.0.tar.gz", hash = "sha256:8572b47aa684357c3e5e10d290ec772c65414114939e3ad2955203e27cd2fcbc", size = 233482, upload-time = "2025-08-18T23:40:21.733Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/09/4e/0ad8585d05312074bb69711b2d81cfed69ce0ae441913d57bf169bed20a7/google_genai-1.52.0.tar.gz", hash = "sha256:a74e8a4b3025f23aa98d6a0f84783119012ca6c336fd68f73c5d2b11465d7fc5", size = 258743, upload-time = "2025-11-21T02:18:55.742Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/41/27/1525bc9cbec58660f0842ebcbfe910a1dde908c2672373804879666e0bb8/google_genai-1.31.0-py3-none-any.whl", hash = "sha256:5c6959bcf862714e8ed0922db3aaf41885bacf6318751b3421bf1e459f78892f", size = 231876, upload-time = "2025-08-18T23:40:20.385Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/66/03f663e7bca7abe9ccfebe6cb3fe7da9a118fd723a5abb278d6117e7990e/google_genai-1.52.0-py3-none-any.whl", hash = "sha256:c8352b9f065ae14b9322b949c7debab8562982f03bf71d44130cd2b798c20743", size = 261219, upload-time = "2025-11-21T02:18:54.515Z" },
 ]
 
 [[package]]
@@ -2509,7 +2509,7 @@ requires-dist = [
     { name = "faker", specifier = ">=36.1.0" },
     { name = "fastapi", marker = "extra == 'desktop'", specifier = ">=0.115.6" },
     { name = "fastapi", marker = "extra == 'server'", specifier = ">=0.115.6" },
-    { name = "google-genai", specifier = ">=1.15.0" },
+    { name = "google-genai", specifier = ">=1.52.0" },
     { name = "granian", extras = ["uvloop", "reload"], marker = "extra == 'experimental'", specifier = ">=2.3.2" },
     { name = "grpcio", specifier = ">=1.68.1" },
     { name = "grpcio-tools", specifier = ">=1.68.1" },