fix(gemini): Add role="user" to function_response content

[Mr-Neutr0n]

Summary

• Adds missing role="user" when constructing ContentType for function responses in Gemini/Vertex AI transformation
• Without this role, multi-turn function calling fails with INVALID_ARGUMENT (400) from the Google API

The Bug

When using Gemini models with multi-turn function calling, LiteLLM fails to include role="user" when constructing the function_response part of the payload.

Error: 400 INVALID_ARGUMENT from Google API

Changes

File: litellm/llms/vertex_ai/gemini/transformation.py

Two locations where tool_call_responses are appended without role:

# Before (line 482)
contents.append(ContentType(parts=tool_call_responses))

# After
contents.append(ContentType(role="user", parts=tool_call_responses))

# Before (line 492)  
contents.append(ContentType(parts=tool_call_responses))

# After
contents.append(ContentType(role="user", parts=tool_call_responses))

Test plan

• Test multi-turn function calling with Gemini models
• Verify no 400 INVALID_ARGUMENT errors

Fixes #20690

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 9, 2026 0:52am

[CLAassistant]

All committers have signed the CLA.

[greptile-apps]

Greptile Overview

Greptile Summary

This PR updates the Gemini/Vertex message transformation so that accumulated tool/function responses are wrapped in a ContentType with role="user" before being appended to contents. This aligns tool result turns with the role expectations of the Google APIs and avoids multi-turn function-calling failures caused by missing roles.

The change is localized to _gemini_convert_messages_with_history() in litellm/llms/vertex_ai/gemini/transformation.py, affecting only the payload constructed for /generateContent requests.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk.
• The change is a narrow, mechanical adjustment to request payload construction (adding an explicit role on tool response content) with no side effects outside Gemini/Vertex message transformation. It fixes a concrete API contract issue and does not alter control flow or data beyond the required field.
• No files require special attention

Important Files Changed

Filename	Overview
litellm/llms/vertex_ai/gemini/transformation.py	Adds explicit role="user" on ContentType entries created for accumulated tool/function responses to satisfy Gemini/Vertex multi-turn function-calling requirements.

Sequence Diagram

sequenceDiagram
  participant Client as LiteLLM Client
  participant Tx as gemini/transformation.py
  participant Gemini as Vertex/Gemini API

  Client->>Tx: _transform_request_body(messages, model,...)
  Tx->>Tx: _gemini_convert_messages_with_history(messages)
  loop iterate messages
    Tx->>Tx: convert_to_gemini_tool_call_invoke(assistant_msg)
    Tx->>Tx: convert_to_gemini_tool_call_result(tool/function msg)
    Tx->>Tx: contents.append(ContentType(role="user", parts=tool_call_responses))
  end
  Tx->>Gemini: POST /generateContent { contents: [...] }
  Gemini-->>Client: 200 response (no INVALID_ARGUMENT on missing role)

Loading

[greptile-apps]

_{1 file reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[krrishdholakia]

@Mr-Neutr0n please add a unit test for this

[Mr-Neutr0n]

I]recheck

[Mr-Neutr0n]

recheck

[Mr-Neutr0n]

recheck

[Mr-Neutr0n]

@krrishdholakia Added unit tests in the second commit (f3dcc25). The tests verify that:

1. function_response ContentType entries have role="user" when converting tool messages
2. Multiple consecutive tool responses are correctly grouped with role="user"
3. Non-tool messages are unaffected

All CI checks pass (lint, unit-test, MCP tests, Helm tests). The mock test suite is running.

[Mr-Neutr0n]

recheck

[Mr-Neutr0n]

@krrishdholakia Are we safe to merge?