smart-compact

You can compress long LLM chat histories by analyzing session context to produce concise summaries, preserving key decisions while estimating token usage to optimize retention within the context window.

Session history compression reduces long chat histories into concise, actionable summaries. You need it when multi-type sessions, including code blocks and planning discussions, exceed token limits and require optimized retention.

You can estimate token usage by running a deterministic dry-run workflow, which calculates the retention ratio and projected token savings before performing the actual session summarization.

Yes, the summarization process handles mixed session types by applying selective retention rules to extract key decisions from both natural language dialogue and code blocks simultaneously.

No, the dry-run and compress workflow operates using local Python scripts and a minimax-like summarization service, requiring no external dependencies to execute session compression.

The best way to retain key decisions is applying selective retention rules during summarization, which automatically identifies and preserves critical planning outcomes while discarding redundant conversational history.