The context window fills as you chat. These commands show what is in it, shrink it, and keep facts across sessions.
See what the context holds
Run /context for a breakdown of the window: model, tokens used and free, how much goes to the system prompt, tools, memory files, skills, and your messages, and where you stand against the automatic compression thresholds. Run /context detail for a per-item list with each tool, memory file, and skill's token count.
Compress a long conversation
Run /compress (or /compact, /summarize) to replace the history with a summary and free up the window. Add instructions to steer the summary, for example /compress keep the migration plan. Instructions longer than 2000 characters are truncated.
/compress-fast makes no model call. It strips old tool outputs and thinking parts, so it is quick and costs nothing. The trade-off is a coarser cut.
Start a clean session
Run /clear (or /new, /reset) to drop the history, free the window, and start a new session. The command refuses while background tasks are running, so stop them first.
Catch up on the session
Run /recap for a one-line overview of the current session.
Give the project a memory
Run /init to analyse the project and write a MATILDA.md. New sessions load that file into context, so keep it short and useful. If a non-empty MATILDA.md already exists, /init asks before overwriting it.
Remember things for next time
Run /remember <text> to save a durable fact for future sessions, and /forget <text> to remove matching auto-memory entries.
Run /memory to open the Memory manager: toggle auto-memory and auto-dream, and open your user memory file, this project's memory file, or the auto-memory folder to edit by hand. Run /dream to consolidate auto-memory topic files on demand.