Chat Completions Api

Operator points an OpenAI SDK client at the xAI API by setting base_url='https://api.x.ai/v1' and model='grok-4'. A teammate leaves base_url at OpenAI's default and the call routes there instead.

xAI exposes an OpenAI-compatible /v1/chat/completions endpoint at https://api.x.ai/v1. Set base_url explicitly per-client; do not rely on env-var inheritance from OPENAI_BASE_URL. Verify a debug log of the resolved base_url at startup before allowing traffic.

Agent sets max_tokens=256 for a long-form Grok answer. Response returns with choices[0].finish_reason='length' and a visibly truncated last sentence.

Detect finish_reason='length' and either (a) surface to caller as partial, or (b) issue a continuation request with the truncated assistant message appended and a 'continue' user turn. Never present the truncated answer as complete. Distinguish from finish_reason='stop'.

Agent places a 2k-token system prompt followed by 40 turns of user/assistant history. System instructions are repeated as a user reminder every 5 turns 'to keep Grok on-task.'

Place system instructions exactly once in messages[0] with role='system'. Do not duplicate as periodic user reminders — that mutates the user channel and degrades instruction-following measurement. If reinforcement is needed, include it in the system message.