When using Gemini 3 flash preview and when it does...
# agentsq
quick-gold-51494
12/17/2025, 10:42 PMWhen using Gemini 3 flash preview and when it does tool calls with arguments, I keep getting this error, despite having thinking budget enabled. Anyone else facing this?
Copy code
raise ClientError(status_code, response_json, response)
google.genai.errors.ClientError: 400 Bad Request. {'message': '{\n "error": {\n "code": 400,\n "message": "Function call is missing a thought_signature in functionCall parts. This is required for tools to work correctly, and missing thought_signature may lead to degraded model
performance. Additional data, function call `default_api:find_patient_by_name_dob` , position 13. Please refer to <https://ai.google.dev/gemini-api/docs/thought-signatures> for more details.",\n "status": "INVALID_ARGUMENT"\n }\n}\n', 'status': 'Bad Request'}quick-gold-51494
12/17/2025, 10:46 PMDecember 17, 2025 (TODAY!) - There was a commit that adds:"Gemini 3 support with thinking_level and thought_signature" (#4027)Your Version vs Latest VersionStatus*You have*: 1.3.6❌ Missing Gemini 3 thought signature fix*Latest*: 1.3.8✅ Has the fix (released today!) Solution: Upgrade LiveKit Plugin
Copy code
pip install --upgrade livekit-plugins-google Copy code
livekit-plugins-google>=1.3.8quick-gold-51494
12/17/2025, 10:49 PMThis didn't seem to fix the issue too
quick-gold-51494
12/17/2025, 10:49 PM Copy code
14:48:38.943 ERROR livekit.agents Error in _llm_inference_task
Traceback (most recent call last):
File "/Users/krishnachoudhary/Documents/errands-ecw-new-endpoints-identity/agents/.venv/lib/python3.11/site-packages/livekit/plugins/google/llm.py", line 427, in _run
raise APIStatusError(
livekit.agents._exceptions.APIStatusError: no response generated (status_code=-1, request_id=41325ae23a79, body=None, retryable=False)
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/Users/krishnachoudhary/Documents/errands-ecw-new-endpoints-identity/agents/.venv/lib/python3.11/site-packages/livekit/agents/utils/log.py", line 16, in async_fn_logs
return await fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/krishnachoudhary/Documents/errands-ecw-new-endpoints-identity/agents/.venv/lib/python3.11/site-packages/opentelemetry/util/_decorator.py", line 71, in async_wrapper
return await func(*args, **kwargs) # type: ignore
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/krishnachoudhary/Documents/errands-ecw-new-endpoints-identity/agents/.venv/lib/python3.11/site-packages/livekit/agents/voice/generation.py", line 130, in _llm_inference_task
async for chunk in llm_node:
File "/Users/krishnachoudhary/Documents/errands-ecw-new-endpoints-identity/agents/src/agents/base.py", line 169, in llm_node
async for chunk in Agent.default.llm_node(
File "/Users/krishnachoudhary/Documents/errands-ecw-new-endpoints-identity/agents/.venv/lib/python3.11/site-packages/livekit/agents/voice/agent.py", line 417, in llm_node
async for chunk in stream:
File "/Users/krishnachoudhary/Documents/errands-ecw-new-endpoints-identity/agents/.venv/lib/python3.11/site-packages/livekit/agents/llm/llm.py", line 350, in __anext__
raise exc # noqa: B904
^^^^^^^^^
File "/Users/krishnachoudhary/Documents/errands-ecw-new-endpoints-identity/agents/.venv/lib/python3.11/site-packages/livekit/agents/llm/llm.py", line 177, in _traceable_main_task
await self._main_task()
File "/Users/krishnachoudhary/Documents/errands-ecw-new-endpoints-identity/agents/.venv/lib/python3.11/site-packages/livekit/agents/llm/llm.py", line 196, in _main_task
return await self._run()
^^^^^^^^^^^^^^^^^
File "/Users/krishnachoudhary/Documents/errands-ecw-new-endpoints-identity/agents/.venv/lib/python3.11/site-packages/livekit/plugins/google/llm.py", line 472, in _run
raise APIConnectionError(
livekit.agents._exceptions.APIConnectionError: gemini llm: error generating content no response generated (status_code=-1, request_id=41325ae23a79, body=None, retryable=False) (body=None, retryable=False)
{"pid": 82091, "job_id": "AJ_t4FimzNCCAC6", "room_id": "RM_i4qQp5ipzaPf"}
Traceback (most recent call last):quick-gold-51494
12/17/2025, 11:22 PMThe Problem
Gemini 3 has a feature called "thinking" - it internally reasons before responding. When it makes a function/tool call, it includes an encrypted blob called a
thought_signaturethought_signature Copy code
┌─────────────────────────────────────────────────────────────────┐
│ TURN 1: User says "Book appointment for tomorrow" │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ Gemini 3 Response: │
│ - function_call: check_availability │
│ - call_id: "abc123" │
│ - thought_signature: <encrypted blob> ◄── WE MUST SAVE THIS │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ Our code executes check_availability, gets result │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ TURN 2: Send back to Gemini 3 with history: │
│ │
│ - Original user message │
│ - function_call: check_availability │
│ call_id: "abc123" │
│ thought_signature: <encrypted blob> ◄── MUST INCLUDE THIS │
│ - function_response: {available: true} │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ Gemini 3 validates signature ✓ and continues conversation │
└─────────────────────────────────────────────────────────────────┘ Copy code
Response comes in
│
▼
┌──────────────────────────────┐
│ _parse_part() sees: │
│ function_call.name │
│ function_call.id = "abc123"│
│ thought_signature = <blob> │
└──────────────────────────────┘
│
▼
┌──────────────────────────────┐
│ Store in dictionary: │
│ _thought_signatures = { │
│ "abc123": <blob> │
│ } │
└──────────────────────────────┘ Copy code
Building next request
│
▼
┌──────────────────────────────┐
│ to_chat_ctx() builds history │
│ Sees function_call "abc123" │
│ Looks up in thought_signatures│
│ Attaches signature to payload│
└──────────────────────────────┘
│
▼
┌──────────────────────────────┐
│ Final payload includes: │
│ { │
│ "function_call": {...}, │
│ "thought_signature": <blob>│
│ } │
└──────────────────────────────┘l
lively-winter-88668
12/18/2025, 10:04 AMSorry, I'm not sure I follow... this is the pull request you're referring to: https://github.com/livekit/agents/pull/4027
What Our Patches DoAre you the author of the pull request?
atest: 1.3.8✅ Has the fix (released today!)I don't think is correct, I don't see it listed in https://github.com/livekit/agents/releases/tag/livekit-agents%401.3.8
2 Views