ai-memory

About

Cross-agent, cross-machine memory and handoffs for coding CLIs

ai-memory runs as one self-contained Rust binary and a server you host on your laptop, a homelab box or anywhere else. Lifecycle hooks silently capture prompts, tool calls and session boundaries, sanitize them, and consolidate each session into readable markdown pages in the project's wiki. The next session, in any supported agent on any machine, gets a short brief and can search everything. Handoffs are typed: each has an owner and is claimed exactly once. A small cross-project profile (your package manager, test layout, habits) follows you to new repos. The default path makes zero LLM calls; memory decays on a tunable schedule, and an optional idle-time "dream" pass can merge cold notes if you configure a model. Multi-user auth and an audit log are built in. Supports Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Grok, Devin, Kiro and more. MIT, about 10k GitHub stars, v3.0 shipped this week. Repo: https://github.com/akitaonrails/ai-memory.

Features

• Shared memory across 20+ agent harnesses
• Typed handoffs between agents, claimed exactly once
• Plain markdown in a git-backed wiki (open it in Obsidian)
• Works across machines and teams with multi-user auth
• Zero LLM calls on the default path
• Memory decay, dedup and optional idle-time consolidation

Install