Time-machine
Rewind the corpus and ask what was known on any date.
Local-first memory for AI
Persistent semantic memory for every agent. Private, local, and yours.
Sessions end. Knowledge should not.
memo carries durable decisions, facts, and preferences into the next session before your agent answers.
A fresh agent reconstructs yesterday from scratch.
A memo-aware agent starts with what already matters.
One local loop
Durable knowledge becomes readable Markdown, a hybrid index, and precise context for any MCP-aware agent.
Capture durable knowledge
Index Markdown locally
Recall it in the next session
More than a vector store
memo knows when knowledge changed, where it came from, and which agent needs it.
Rewind the corpus and ask what was known on any date.
Find stale decisions and resolve conflicts without erasing history.
Extract durable insights from real work without constant remember commands.
Claude Code, Codex, Cursor, Devin, and other MCP clients share one memory.
Smaller context. Stronger continuity.
A compact MCP surface and tight recall budget reduce repeated model work while public project activity stays verifiable.
Public data updated:
Runs on your machine
Use MLX on Apple Silicon or the CPU backend on Linux. Your prompts and memories stay local.
curl -fsSL https://raw.githubusercontent.com/jagoff/memo/master/install.sh | bashpipx install "mlx-memo[cpu]"Why memo
| Capability | memo | Cloud memory | Vector store |
|---|---|---|---|
| Local by default | Yes | No | Sometimes |
| Time travel | Yes | Rare | No |
| Contradiction handling | Yes | Partial | No |
| Cross-agent recall | Yes | Partial | Custom |
Your agents already learn
Install memo once and give every session a durable starting point.