Cheap small-model text completion for agent sub-tasks: classify, extract, summarize, rewrite. MUST be invoked when an agent needs a short LLM answer without paying flagship rates. Prompt <=8000 chars, maxTokens <=1024, json:true forces a JSON object. Do NOT use for long-form generation or tool calling. Settles $0.003 USDC; no charge on failure; receipts at /api/ledger.