fix(ollama): forward tool_calls and tool_call_id in transform_request

[mverrilli]

Summary

• transform_request translated tool_calls on assistant messages to OllamaToolCall format but never copied them into the outgoing OllamaChatCompletionMessage — Ollama received {role: assistant, content: ''} with no tool_calls
• The model had no record of having made a tool call and re-issued the same call on every turn, causing an infinite loop
• tool_call_id on role: tool messages was also silently dropped; Ollama uses this field to resolve the tool function name from conversation history
• Added tool_call_id to OllamaChatCompletionMessage TypedDict

Fixes #26094 (reported via ollama/ollama#15719)

Test plan

• TestOllamaToolCallTransformation::test_transform_request_preserves_tool_calls — asserts tool_calls survive the transform on assistant messages
• TestOllamaToolCallTransformation::test_transform_request_forwards_tool_call_id — asserts tool_call_id is forwarded on tool response messages
• Full test_ollama_chat_transformation.py suite: 24/24 pass

[CLAassistant]

All committers have signed the CLA.

[greptile-apps]

Greptile Summary

This PR fixes two bugs in the Ollama chat transformation: tool_calls on assistant messages were translated to OllamaToolCall format but never written into the outgoing OllamaChatCompletionMessage, and tool_call_id on role: tool messages was silently dropped. Both fields are now correctly forwarded and OllamaChatCompletionMessage TypedDict has been extended with tool_call_id. Tests cover both regression cases and make no real network calls.

Confidence Score: 5/5

Safe to merge — targeted bug fix with full regression test coverage and no side effects.

The change is minimal and surgical: it corrects two clear omissions (missing tool_calls assignment and dropped tool_call_id) and removes an existing no-op mutation. Both fixes are covered by new unit tests. No security concerns, no backwards-incompatible changes, and no modifications to mocks that could mask regressions.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/llms/ollama/chat/transformation.py	Correctly assigns translated tool_calls to ollama_message and forwards tool_call_id; removed the prior no-op in-place mutation of m.
litellm/types/llms/ollama.py	Adds tool_call_id field to OllamaChatCompletionMessage TypedDict; minimal, correct change.
tests/test_litellm/llms/ollama/test_ollama_chat_transformation.py	Adds two regression tests for the fixed bugs; both are unit tests with no real network calls, following repository conventions.

Sequence Diagram

sequenceDiagram
    participant Client
    participant LiteLLM as LiteLLM (transform_request)
    participant Ollama

    Client->>LiteLLM: messages=[user, assistant(tool_calls), tool(tool_call_id)]
    Note over LiteLLM: Translate tool_calls → OllamaToolCall format
    Note over LiteLLM: Assign new_tools to ollama_message["tool_calls"] ✅ (was missing)
    Note over LiteLLM: Forward tool_call_id to ollama_message ✅ (was dropped)
    LiteLLM->>Ollama: {role:assistant, tool_calls:[…]}, {role:tool, tool_call_id:"…"}
    Ollama-->>LiteLLM: Response (no re-issued duplicate call)
    LiteLLM-->>Client: Final answer

Loading

_{Reviews (3): Last reviewed commit: "fix(ollama): forward tool_calls and tool..." | Re-trigger Greptile}

[roxlukas]

Tested with LiteLLM v1.83.7.rc.1 and Ollama 0.21.2 - problem is still visible

[mverrilli]

@roxlukas My fix is not in https://github.com/BerriAI/litellm/releases/tag/v1.83.7.rc.1 do you mean that you applied my PR on top of that tag and it still is not working?

I just tested it again:

• Pre-fix payload (v1.83.7.rc.1, what you tested) is demonstrably broken:

{"role":"assistant","content":"","images":[]}          ← tool_calls GONE
{"role":"tool","content":"...","images":[]}           ← tool_call_id GONE

Ollama has no record a tool was ever called. The loop is inevitable.

• Transform output matches Ollama's native format (post-fix):

{"role":"assistant","content":"","tool_calls":[{"function":{"name":"get_weather","arguments":{"city":"Warsaw"}}}]}
{"role":"tool","content":"...","tool_call_id":"call_abc"}

Identical call, but with the tools showing.

If you did apply my PR on top of that branch, tell me more about what you are seeing and I can try to reproduce.

I also rebased it.

Thanks

[veria-ai]

Low: Bug fix for Ollama tool call forwarding

This PR fixes a bug where tool_calls and tool_call_id fields were being dropped during request transformation for the Ollama provider. The changes are mechanical — values are passed through typed constructors into a TypedDict that gets JSON-serialized for the Ollama API request. No security concerns.

---

Status: 0 open
Risk: 1/10

Posted by Veria AI · 2026-04-24T22:43:52.062Z

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!