SPI_202 — Conversation title generator
Summarizes a new thread's first user message into a short, human-readable conversation title; runs overlapped with the answer on the first /ask turn.
| ID | SPI_202 |
| Ownership | Internal (not part of your agent’s prompt) |
| Sent when | First turn of a new thread when Auto-title is on and the tenant is not no-store. |
| Flows | classic · agentic |
| Delivered as | standalone-request |
| Model | The tenant's resolved chat model (default gpt-4.1-mini), on the tenant's BYOK key. |
Summarizes a new thread's first user message into a short, human-readable conversation title; runs overlapped with the answer on the first /ask turn.
It is kicked off at request start and runs concurrently with RAG, tools, and the main answer, so it adds ~no latency — the title ships inline on the first /ask response (an optional title field, also emitted on the SSE done event). Later turns return the persisted title for free. Only brand-new threads are titled; resumed threads keep their existing title and a user rename always wins. If Auto-title is off, or the call fails or times out, the title falls back to a truncation of the first user message. No-store tenants are skipped entirely.
In the debug trace
Not part of the assembled agent prompt in the /ask trace — the title generator is a separate internal LLM call kicked off at request start, never joining your agent's prompt.