Gemini 3.5 flash (gemini 3+ models) + Vertex AI + LiteLLM tool calling fails with invalid id fields after thought_signature fix

[spanda1985]

I am using LibreChat (v0.7.9)with LiteLLM as an external proxy and Vertex AI Gemini models.

Previously, with LiteLLM 1.79.x, tool calling failed due to missing thought_signature handling.

After upgrading LiteLLM to 1.87.0-dev.1, the original thought_signature error appears resolved, but now requests fail with a different Vertex AI validation error related to unsupported id fields inside function_call and function_response.

This appears to happen specifically during tool-calling/web-search flows.

The issue now looks like an OpenAI-style tool payload is leaking into the Vertex Gemini payload format.

Previous Error (LiteLLM 1.79.3)

Previously we consistently received:

Something went wrong. Here's the specific error message we encountered:

An error occurred while processing the request: 503 data:
{
"error": {
"message": "litellm.ServiceUnavailableError:
litellm.MidStreamFallbackError:
litellm.BadRequestError:
Vertex_ai_betaException BadRequestError - b'{
"error": {
"code": 400,
"message": "Unable to submit request because function call default_api:web_search in the 4. content block is missing a thought_signature. Learn more: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/thought-signatures\",
"status": "INVALID_ARGUMENT"
}
}'

Current Error (LiteLLM 1.87.0-dev.1)

After upgrading LiteLLM, the previous thought_signature error no longer occurs.

Now requests fail with:

Something went wrong. Here's the specific error message we encountered:

An error occurred while processing the request: 400

litellm.BadRequestError:
Vertex_ai_betaException BadRequestError - b'{
"error": {
"code": 400,
"message": "Invalid JSON payload received.
Unknown name "id" at 'contents[1].parts[1].function_call': Cannot find field.
Invalid JSON payload received.
Unknown name "id" at 'contents[2].parts[0].function_response': Cannot find field.",
"status": "INVALID_ARGUMENT",
"details": [
{
"@type": "type.googleapis.com/google.rpc.BadRequest",
"fieldViolations": [
{
"field": "contents[1].parts[1].function_call",
"description": "Invalid JSON payload received. Unknown name "id" at 'contents[1].parts[1].function_call': Cannot find field."
},
{
"field": "contents[2].parts[0].function_response",
"description": "Invalid JSON payload received. Unknown name "id" at 'contents[2].parts[0].function_response': Cannot find field."
}
]
}
]
}
}'

Sample from gemini resp
LiteLLM:DEBUG[0m: vertex_and_google_ai_studio_gemini.py:3398 - RAW GEMINI CHUNK:

     { 'candidates': [{ 'content': { 'role': 'model', 'parts': [{ 'functionCall': { 'name': 'web_search', 'args': { 'news': True, 'query': 'latest news Europe May 2026' } }, 'thoughtSignature': 'AY89a1990/.......' }] } }], 'usageMetadata': { 'trafficType': 'ON_DEMAND' }, 'modelVersion': 'gemini-3.5-flash', 'createTime': '2026-05-22T00:23:38.013261Z', 'responseId': 'CqIPas1ntPXq2Q_OkYCwCA' }

From the repo i see that this might already been fixed.

If anyone can please help or point out to a patch/fix would really appreciate without upgrading to latest librechat.

Thanks in advance.

[omni-front]

Remove the id fields from function_call and function_response in your payload to match Vertex AI's expected format.

[spanda1985]

And this is only for gemini models? Do we have a patch in libre repo for this?

[danny-avila]

Use the latest version of LibreChat