Grok API: send the encrypted reasoning back every turn

xAI’s best-practices guide says to use the Responses API (Chat Completions is deprecated) and keep the model’s reasoning in the conversation.

Two ways

• Chain requests with previous_response_id (with store on, the default).
• Or store: false and append every output item, including reasoning items with encrypted_content, to the next request’s input.

Common ways code drops it

• Extracting only output_text and re-adding it as a plain message.
• Editing, filtering or reordering output items.
• Rebuilding history in Chat Completions format between turns.

Cheaper too

Give each conversation its own prompt_cache_key and only append. Missing reasoning items are the most common cause of cache misses.