Skip to main content

Source: https://reclaimllm.com/news/auto-summarization-background-task

← Back to News feature Mar 15, 2026

Sessions Now Auto-Summarize After Ingest

New sessions are automatically titled, described, and tagged in the background — no manual trigger required.

Sessions Now Auto-Summarize After Ingest

Sessions ingested into ReclaimLLM now receive a title, description, and tags automatically. Previously this required an explicit API call after ingest. That endpoint still exists, but you no longer need to call it.

What to Expect

After a session is ingested, a background task wakes periodically, picks up any untitled sessions, and generates summaries via Azure OpenAI. The default interval is 5 minutes, so titles typically appear within a few minutes of ingest under normal load. Duplicate sessions — those with identical blob content — are detected and cleaned up in the same pass. Only one copy is kept and summarized; the rest are dropped without making any OpenAI calls.

Limitations

  • Latency is interval-bound, not ingest-bound. A session ingested right after a batch run will wait up to the full interval before being summarized. If you need a title immediately, the on-demand summarize endpoint is still available.
  • Backlog drains linearly. After an outage or first deploy, a large backlog of untitled sessions clears at a fixed batch size per cycle. Expect a few cycles before everything is titled.
  • OpenAI outages delay summaries, not ingests. If Azure OpenAI is unavailable, sessions continue to ingest normally. Summarization resumes automatically once the service recovers — no data is lost.

What’s Next

This is the first version of auto-summarization. The current polling approach is intentionally simple: no new infrastructure, no separate worker process. If tighter latency (sub-minute) or higher ingest volume becomes a requirement, we’ll migrate to a queue-based model. That’s tracked and will be communicated here when the time comes. Feedback welcome — if you see sessions stuck untitled longer than expected, open an issue. ShareShare on X Share on LinkedIn Share on Facebook