[Ar9av]

Problem

prismor elevenlabs connect --model <m> points the agent at the proxy with api_type: chat_completions. Three models can't serve that, so every turn of the call fails at OpenAI. Checked live on 2026-11-07:

Model	Chat completions result
gpt-5-pro	404, "only supported in v1/responses"
gpt-6-astra	400 with a function tool: "Function tools with reasoning_effort are not supported … in /v1/chat/completions"
gpt-6.1-sol	same 400 as Astra

Follow-up to #875, which fixed the reasoning_effort value for these models. Its author noted the tool-calling gap and left it out of scope.

Solution

connect() refuses these models up front, with an error that names models that work. It refuses before any secret is created, any agent is patched, or proxy.json is written. Dated snapshots (gpt-6-astra-2026-09-01) are matched the same way #875 matches them.

Astra and Sol 6.1 are refused even for agents with no tools. Whether ElevenLabs still sends system tools like end_call on those turns isn't verified. A per-agent tool check can be added if someone needs those models without tools.

Files changed

File	Why
prismor/runtime/elevenlabs_cli.py	_CHAT_UNSERVABLE and the check at the top of connect()
tests/test_elevenlabs.py	parametrized refusal test; checks the agent and proxy.json are left untouched
docs/elevenlabs.md	one line under --model

Testing

python -m pytest tests/test_elevenlabs.py tests/test_elevenlabs_parallel.py tests/test_elevenlabs_refusal_text.py tests/test_proxy.py -q

133 passed (macOS).

Conflicts with #875 on import re. Whichever merges second needs a one-line rebase.