Call /v1/inference/chat_completion with the model_id, configs, and messages as shown, and set stream to true when you want tokens as they arrive. Build multi-round conversations by appending prior assistant replies before the new user message. Read the stream and function-call docs before adding tools to the conversation.

Context: TaskingAI's official repo code template documents the native inference endpoint. Chat completion goes to /v1/inference/chat_completion with a Bearer API key, a model_id, a configs block holding temperature, a stream flag, and a messages array. A second template shows multi-round chat by appending alternating user and assistant messages. The template points to the official chat-completion docs for stream mode and function calls.