session-memory

Cross-session memory for AI agents is maintained by reading and writing persistent session files to the file system, storing decisions, mistakes, and context to prevent repetition and enhance learning. You load previous context at session start and process it at the end.

Your AI agent repeats mistakes because it lacks persistent cross-session memory. By recording decisions and discoveries to file system memory files during work, the agent loads this context at the start of a new session to avoid repetitive errors.

The best way to retain context is to use a continuous recording mechanism that logs decisions during work and compacts memory at session end. This creates persistent session summaries that can be loaded when starting a new session.

Yes, you need file system access to manage persistent memory for AI agents. The session memory mechanism requires reading and writing access to store and load session memory files that contain decisions, mistakes, and repository context.

To process and compact memory at the end of an AI agent session, run the session-end command. This compacts memory, creates session summaries, and verifies records to ensure that context is properly retained for future sessions.