Long conversations & recall
Chats that outgrow the model's context keep everything, and one chat can read another.
Long conversations
Every model has a limit to how much conversation it can read at once. When a chat approaches it, Actovia compacts the older part into summaries — a card says Compacting conversation history…, then Compacted 24 messages · saved ~18k tokens. Nothing is thrown away: the original messages stay stored, and the model can search them and open any summary back up to the full text when you ask about something said long ago.
The newest exchanges are always sent whole. Under Settings → Memory → Context management:
- Lossless context management — on by default.
- Compaction threshold — how full the context gets before compacting, 50–95% (default 75%).
- Fresh tail size — how many of your latest messages, with everything that answered them, are never compacted (2–24, default 6).
Across chats
A chat can read another one when you hand it over. In the sidebar, open a chat's menu and choose Copy conversation ID, then paste the ID into another chat with a question — "what did we decide in this one?". The model looks that conversation up, reads what it needs, and says which chat it drew on.
This is not long-term memory: nothing is copied anywhere, and a chat only reads another when you give it the ID.
