Thread TL;DR on demand (P12-M)

Summarize long reply threads without scrolling the full history. Reuses the same AI path as reply suggestions (P12-D) and daily digest (P12-F).

Usage

API

POST /messages/{messageId}/summary
Authorization: Bearer <jwt>
Content-Type: application/json

{ "roomId": "room_support" }

Response

{
  "summary": "- Customer asked about refund policy\n- Agent offered store credit\n- Follow up Friday",
  "rootMessageId": 42,
  "messageCount": 5,
  "truncated": false
}

SDK

const { summary } = await client.summarizeThread(messageId, roomId);

Dashboard

Root messages with at least two replies show a TL;DR button in the agent room chat.

Behavior

Resolve thread root by walking parent_id up from the anchor message.
Collect root + nested replies (BFS, max 40 messages).
Require at least 3 messages in the thread (thread_too_short otherwise).
Call OpenAI-compatible chat completions via AI_BASE_URL.

Limits

Control	Default
Min thread size	3 messages
Max thread size	40 messages
Quota metric	`agent_invocations`
Rate limit	`RATE_LIMIT_THREAD_SUMMARY_PER_MINUTE` (10/min)
Model	`AI_SUMMARY_MODEL` → `AI_SUGGEST_MODEL` → `AI_MODEL`

Errors

Code	HTTP	Meaning
`thread_too_short`	400	Fewer than 3 messages in thread
`message_not_found`	404	Invalid message id for room
`quota_exceeded`	402	Monthly AI quota
`ai_not_configured`	503	Missing `AI_BASE_URL`