SUMMARIZE — condense conversation history to free context window space
/summarize /compact
/summarize (alias: /compact) asks the active model to produce a concise summary of the current conversation, then replaces the entire history with:
This frees up the context window so you can continue a long session without hitting the model’s token limit or degrading response quality from a very-full context.
The summary is generated by the same model you are currently using. Small or instruction-poor models may produce lower-quality summaries. If the summary is empty (e.g. the model refused or failed), history is left unchanged.
At least two non-system messages must exist in history; otherwise the command does nothing.
/compact is an exact alias for /summarize — both names work identically.
harvey > /status
Tokens: ~98 000 / 131 072 (74%)
harvey > /summarize
History condensed to 847 chars.
harvey > /status
Tokens: ~312 / 131 072 (0%)
System prompt — re-injected automatically. Pinned context — re-injected as the first user message. RAG state — unaffected. Recording — the running session continues; summary is not written back to the Fountain file.
/context add TEXT — pin a summary you compose manually /clear — discard history entirely (no summary generated) /status — check context window usage before and after /help context /help clear