Answer on an OpenAI key from the Agno Bot, not only on Anthropic or an endpoint

[kevin9327]

What this changes

Picked with an OpenAI key, the Agno Bot fails every run before a request leaves the container.

Agno's LiteLLM model sends temperature=0.7 and top_p=1.0 with every request (agno/models/litellm/chat.py, get_request_params). LiteLLM refuses both for a reasoning model, and gpt-5.5 is one. It is also what the harness gets on that choice: the desktop removes BOT_MODEL for an OpenAI key, so Compose passes its default, gpt-5.5. The run then ends in:

litellm.UnsupportedParamsError: gpt-5.5 only supports logprobs, top_p, top_logprobs when reasoning_effort resolves to 'none', ...

With only the temperature, LiteLLM refuses that too (gpt-5.5 doesn't support temperature=0.7 while reasoning is active).

The model is now built with request_params={"drop_params": True}, set on this one client only, so LiteLLM drops a parameter the model does not take instead of refusing the run. #533 fixed the LlamaIndex Bot the same way, for the temperature LlamaIndex sends. Measured against the fake provider below, the request for gpt-5.5 now carries neither parameter. A compatible endpoint's local-model still gets both, because LiteLLM does not know that model and drops nothing.

Left alone: the Anthropic key and OpenAI-compatible endpoint choices already worked, and every other harness.

Where it runs

• New state that outlives a request? None. One keyword argument on the model the module builds at import.
• What happens on the second replica? The same. Each harness process builds the same client from the same environment.
• Anything serialised? No.
• Anything fanned out to a browser? No.
• New listener, port, or schedule? No. The CI step added here runs beside the three harness suites already in python-harness.

Boundary and audit

• Every acting call still goes through the gateway: nothing on the server changes, and the harness still refuses a request without the server's token.
• New refusals and new failures each write a row: no new refusal. What gets wider: a parameter the model rejects is now dropped without a word instead of failing the run, the same trade Reach the model the setup screen chose from the LlamaIndex Bot, not only an OpenAI one #533 made for LlamaIndex.
• Nothing new is trusted from the client.

Changelog

• A line in CHANGELOG.md under Unreleased.

Proof

• New agent-agno/tests/test_main.py follows agent-llamaindex/tests/test_main.py. It starts a local fake provider serving /v1/chat/completions and /v1/messages, imports the harness under the environment the desktop writes for each of the three choices, and posts an AG-UI run to /agui. It asserts HTTP 200, RUN_FINISHED, no RUN_ERROR, and that the provider got the chosen model on the chosen route.
• Linux (ubuntu-latest, Python 3.12, a throwaway workflow on my fork running the same commands as the new CI step). This branch: 3 passed. With agent-agno/src checked out from main (2d09a08): 1 failed, 2 passed. The failure is [an OpenAI key], with the UnsupportedParamsError above in the stream.
• Windows, Python 3.12: the same results.
• Versions: agno 3.0.10, litellm 1.101.0.
• Not run: a real OpenAI or Anthropic key, and the published image.

This adds a step at the end of python-harness and an entry at the top of Unreleased, the same places other open PRs touch. If one of them lands first this will need a rebase, and I'll do it.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.