OpenAI-compatible & custom webhook

If your agent is not a Microsoft backend, MeetCrew still fronts it. Use the OpenAI-compatible path for any chat-completions endpoint, or the custom webhook for anything else. Your backend reasons; MeetCrew speaks its reply.

OpenAI-compatible endpoints

The OpenAI-compatible backend accepts any endpoint that implements the OpenAI chat-completions API — OpenAI directly, Azure OpenAI, models exposed through a compatibility layer, or a model you host yourself at a public HTTPS address. MeetCrew refuses private, loopback and link-local addresses, so a model reachable only inside your own network cannot be bound. Bind the endpoint to a teammate deployment and authenticate with an API key. MeetCrew’s teammate-side tools are offered through the standard tools parameter, so your agent can call them during a turn. MeetCrew streams the reply, and retries once a call that cannot connect or that answers 5xx or 429.

Custom webhook

The custom webhook is the escape hatch for backends that fit no other category. You supply an HTTPS endpoint that implements the MeetCrew agent contract: it receives the request as JSON and returns the reply as JSON. Requests are authenticated with an HMAC signature over the request body, using a shared secret configured per deployment. A webhook is not streamed: MeetCrew calls it once, without retrying, and speaks the full reply it returns. The same address rules apply as for an OpenAI-compatible endpoint.

How a turn works

When the teammate is addressed by name, MeetCrew calls your endpoint with the meeting context and speaks the reply it returns. An empty reply produces silence, not an invented answer.

Credentials are write-only

You enter API keys and webhook shared secrets once in the admin console. MeetCrew keeps them in a dedicated Azure Key Vault and never shows them again.