Send an authenticated request to your TaskingAI service’s `/v1/chat/completions` endpoint. Use the TaskingAI model ID in `model` and supply the conversation in `messages`. Stateless multi-turn calls include the prior turns the model needs; a new request does not recover them from the previous call.

For the OpenAI Python client, configure `base_url` as the service host plus `/v1`, then call `client.chat.completions.create`. Set `stream=True` and consume the returned chunks when incremental output is needed. Check one complete response or stream against a short known prompt before integrating it. Other parameter support depends on the deployed compatibility implementation; this entry does not generalize assistant-application presets to model invocations.