ChatHub appears to rely on different logic then the normal AI Agent workflow as when you configure a custom base url you cannot disable the Response API setting making nearly all other OpenAI compatible providers broken for Chat Hub.
This may be a request-shape mismatch rather than a key problem.
A no-secret way to narrow it down: compare what ChatHub sends versus the normal AI Agent path:
- Does ChatHub force `/v1/responses` while your OpenAI-compatible provider only supports `/v1/chat/completions`?
- Does the custom base URL end at `/v1`, not at a deeper endpoint?
- Is the same model name working in the AI Agent node but failing only in ChatHub?
- What is the redacted error category: 404/model-not-found, unsupported endpoint, schema error, or timeout?
Please do not share API keys or raw logs. If the issue is that ChatHub cannot disable Responses API, that would explain why many OpenAI-compatible providers break while the normal AI Agent workflow still works.