MapleAI Autonomous Agent Execution

Choose this endpoint when the user's request requires multiple sequential reasoning steps, tool use (web search, code execution, URL fetching, data analysis, or calculation), or autonomous decision-making across steps — rather than a single-shot LLM completion. Prefer oss-20b for cost-sensitive tasks and gpt-6-sol for complex, high-accuracy needs. Use stream=true when you need to show progress to a user in real time. This is the right endpoint over a simple chat completion when the task cannot be answered in one inference call and requires tool orchestration.

• Task exceeds max_steps limit (20 max) — agent stops and returns partial result
• code_exec quota exceeded — max 3 sandboxed code calls per task; additional calls are blocked
• Invalid tool name in tools array — returns validation error
• Model enum mismatch — only 'agents/oss-20b' or 'agents/gpt-6-sol' accepted
• Payment failure — x402 payment not accepted or insufficient USDC balance
• SSE stream disconnected mid-task — partial steps may be lost
• Task too ambiguous for agent to make progress — returns low-confidence or empty answer
• fetch_url tool blocked by target site — step fails with fetch error