persistent-memory

To keep AI context across sessions, you use a portable, file-based memory protocol that writes decisions and history to a MEMORY.md file and session logs, allowing any file-reading AI agent to resume work seamlessly without losing prior context.

A file-based memory protocol is a standardized structure using a .planning directory to store a MEMORY.md brain, session logs, and handoff notes, enabling cross-session continuity for any AI agent that can read and write files.

You initialize cross-session memory by running a memory initialization command to bootstrap a .planning directory, which creates the MEMORY.md file and session log structure for the AI agent to read at the start of each new session.

Yes, you can use this memory protocol with any AI agent that has the ability to read and write files, because it relies on a standardized file-based structure rather than a specific framework or proprietary API integration.

You manage file size by using the protocol's lightweight compression and archiving features, which keep the MEMORY.md file under a reasonable size while preserving historical session logs and decision records in the .planning directory.

The best way to handle AI agent handoffs is by writing standardized handoff notes within the .planning directory, which the next session reads alongside MEMORY.md to restore context, review past decisions, and guide execution.