[FEATURE]: Silent/background compaction — don't stream summary output to the terminal #9013

Open
opened 2026-02-16 18:11:24 -05:00 by yindo · 0 comments
Owner

Originally created by @Rohansguliani on GitHub (Feb 10, 2026).

Originally assigned to: @thdxr on GitHub.

Feature hasn't been suggested before.

  • I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

When auto-compaction triggers, the entire summarization process streams visibly into the chat window. The user has to sit and watch the summary text render token-by-token before they can continue working. This is disruptive and slow — especially since compaction uses the same primary model as the active session by default.

Current behavior:
The compaction summary streams into the chat window just like a normal assistant response. This interrupts the user's flow, is slow (uses the primary model), and adds visual noise — the summary is internal context for the AI, not something the human needs to read in real-time.

Expected behavior:
Compaction should run silently in the background. The user should see at most a brief indicator (e.g., "Compacting context..." → "Done"), not the full streamed output.

Suggestion:

  • Stream compaction in the background without rendering output to the chat
  • Show a minimal status indicator (spinner, progress bar, or one-liner like "Context compacted")
  • Optionally, make this configurable (e.g., compaction.silent: true) for users who want to watch it

Not a duplicate of flagged issues:

  • #6228 is about disabling per-message summary generation (small summaries, not compaction)
  • #11930 is about configurable compaction threshold and model selection
  • #12479 is about token waste from re-summarizing already-summarized messages

This issue is specifically about the UX: compaction output should not stream visibly into the user's chat window.

Environment: opencode 0.0.0-dev-202602020741, macOS (desktop/web app)

Originally created by @Rohansguliani on GitHub (Feb 10, 2026). Originally assigned to: @thdxr on GitHub. ### Feature hasn't been suggested before. - [X] I have verified this feature I'm about to request hasn't been suggested before. ### Describe the enhancement you want to request When auto-compaction triggers, the entire summarization process streams visibly into the chat window. The user has to sit and watch the summary text render token-by-token before they can continue working. This is disruptive and slow — especially since compaction uses the same primary model as the active session by default. **Current behavior:** The compaction summary streams into the chat window just like a normal assistant response. This interrupts the user's flow, is slow (uses the primary model), and adds visual noise — the summary is internal context for the AI, not something the human needs to read in real-time. **Expected behavior:** Compaction should run silently in the background. The user should see at most a brief indicator (e.g., "Compacting context..." → "Done"), not the full streamed output. **Suggestion:** - Stream compaction in the background without rendering output to the chat - Show a minimal status indicator (spinner, progress bar, or one-liner like "Context compacted") - Optionally, make this configurable (e.g., `compaction.silent: true`) for users who want to watch it **Not a duplicate of flagged issues:** - #6228 is about disabling per-message summary generation (small summaries, not compaction) - #11930 is about configurable compaction threshold and model selection - #12479 is about token waste from re-summarizing already-summarized messages This issue is specifically about the **UX**: compaction output should not stream visibly into the user's chat window. **Environment:** opencode 0.0.0-dev-202602020741, macOS (desktop/web app)
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: anomalyco/opencode#9013