When a chat grows long or is interrupted

A long conversation can reach the selected model's context limit, while an interruption can leave a reply unfinished. These are different situations: compaction summarizes older context to make room; resuming after an interruption uses the work that actually completed. Neither operation restores deleted details or undoes actions already taken.

Make room for more work

Automatic compaction is enabled by default and normally starts at 80% of the model context window. In your Account → Models, you can turn Automatically shorten long conversations on or off and adjust its threshold; Compaction per model lets you set a separate threshold for a model. To summarize earlier history yourself, send /compact in chat or CLI; add a short instruction about details to preserve, for example /compact Keep the decisions and test results. If nothing can be summarized, the command does nothing. If it fails, Elowen reports the failure rather than claiming the context was shortened. Long-term memories are not deleted.

Continue after an interruption

If a reply stops, inspect its final transcript row: it can show that you stopped it, an error occurred, or work reached a limit. The text already produced remains in the conversation. Send a follow-up to ask Elowen to continue from that result or redirect it; do not assume tools or external actions that had not completed were performed.

Check the outcome

If an interruption occurred around a restart, Elowen resumes supported work from its saved recovery state; it may instead park or end work that cannot safely resume. Check the transcript for the final status before repeating actions. A failed send reports the problem, so verify whether the message was delivered before you try again. See Brain & Chat for history and Work Modes & Steering for guidance during an active run.