Roadmap: Large group chats

The situation

Large chat rooms can dominate search, embeddings and relationship statistics even when I only read them occasionally. I want to keep their text without treating every room as part of my active conversations.

What works today, and why it is not enough

Participant-based media limits avoid downloading attachments from large rooms. They do not reduce text search noise or stop those messages from contributing to embeddings and people data. Deleting the archive would lose useful history.

Ask

Add a conversation policy that keeps small or recently active rooms full and otherwise archives large or unknown-size rooms as cold. Cold text should be reachable through explicit lexical search, with no contribution to embeddings, default discovery or people analytics. Allow exact conversation overrides.

Keep destructive retention separate: explicit stub/skip choices, a local preview by default with clear counts and explicit --yes to apply, and no provider deletes. Preserve rosters and make policy changes recoverable through sync or explicit export reimport. Existing attachment limits should continue to apply.

Series

1. Repair archived conversation rosters: #1295.
2. Add the tier model and exact conversation overrides: #1296.
3. Archive inactive rooms as cold across ingestion, search and people features: #1361.
4. Add explicit stub/skip retention and preview-first local history purge.