How do you stop session intent

[nomie7]

how to stop streaming of session intent to make it faster?

[WillemJiang]

The streaming behavior in DeerFlow is controlled at the API endpoint level, not through a global config toggle.

Using non-streaming endpoints

The backend provides three run modes:

┌────────────────────────────────────┬─────────────────────────────────────────────┐
│              Endpoint              │                  Behavior                   │
├────────────────────────────────────┼─────────────────────────────────────────────┤
│ POST /api/threads/{id}/runs/stream │ Streams events via SSE (real-time)          │
├────────────────────────────────────┼─────────────────────────────────────────────┤
│ POST /api/threads/{id}/runs/wait   │ Blocks until completion, returns final JSON │
├────────────────────────────────────┼─────────────────────────────────────────────┤
│ POST /api/threads/{id}/runs        │ Fire-and-forget background run              │
└────────────────────────────────────┴─────────────────────────────────────────────┘

If you're calling the API directly, use the /wait endpoint instead of /stream — it returns the complete result in one response.

Frontend

The Next.js frontend always uses streaming (via LangGraph SDK's useStream hook). There's no built-in toggle to switch it to non-streaming. You'd need to modify frontend/src/core/threads/hooks.ts to call a /wait
endpoint instead.

Speed tips

If your goal is faster perceived response time:

• Use flash mode — disables extended thinking (thinking_enabled: false), significantly reducing latency
• Disable subagents — set subagent_enabled: false in the runtime context
• Lower reasoning_effort — set to low for faster (less thorough) responses

[nomie7]

thanks will try this