Persistent Memory

Persistent memory for AI coding agents that stays in your repo.

Memory gives coding agents a local project memory layer for product intent, architecture decisions, repo conventions, workflows, and gotchas. Agents load a task-focused context pack before work and save durable knowledge after meaningful changes.

What this solves.

Cross-session context Stop losing project decisions between agent sessions.

Instead of rebuilding context from chat history, Memory stores reusable project knowledge as local, reviewable files that future coding agents can load on demand.

Repo-native storage Persistent does not have to mean hosted.

Memory keeps canonical knowledge under `.memory/`, so teams can inspect changes, review them in Git, and repair stale facts instead of trusting an opaque memory service.

Focused retrieval Load the right memory, not every note.

`memory load` compiles a small context pack for coding, debugging, review, architecture, or onboarding work, reducing token noise while keeping important constraints active.

Continue with adjacent project-memory workflows for AI coding agents, AI coding assistants, and MCP-capable tools.