memo — local memory for AI agents
Registers memo as an MCP server in VS Code's
agent mode, so Copilot (and any MCP-aware chat) can remember decisions, facts,
and preferences across sessions — stored locally, on your own machine.
- 🧠 Long-term memory — hybrid
sqlite-vec + BM25 retrieval with reranking, a
knowledge graph, and temporal ("time-machine") queries.
- 🍎 Local & private — MLX embeddings on Apple Silicon, CPU on Linux. Markdown
files on disk are the source of truth. No cloud, no API keys.
- 🔌 Zero config — the extension contributes the
memo MCP server automatically.
Open agent mode, pick memo, and its tools appear.
Requirements
memo runs as a local process. By default the extension launches it with:
uvx --from mlx-memo memo-mcp
So you need uv installed
(uvx ships with it). uvx fetches the mlx-memo package from PyPI on first run — no
manual pip install needed.
Already installed memo yourself (pipx install mlx-memo or uv tool install mlx-memo)?
Turn on memo: Use Installed Binary in settings and the extension will run your
memo-mcp directly.
Usage
- Install this extension.
- Open the Chat view and switch to Agent mode.
- Click the tools icon → memo should be listed. Enable it.
- Ask the agent to remember or recall something — memo persists it to disk.
Run memo: Setup & Requirements from the Command Palette for install help.
Settings
| Setting |
Description |
memo.useInstalledBinary |
Run an already-installed memo-mcp instead of uvx --from mlx-memo. |
memo.dataDir |
Directory for memo's store and index (MEMO_DATA_DIR). Empty = default. |
memo.vaultPath |
Store memories under an Obsidian vault (MEMO_VAULT_PATH + MEMO_MEMORIES_IN_VAULT). |
About
memo is open source (MIT) and distributed on PyPI as
mlx-memo. Source, issues, and full docs:
https://github.com/jagoff/memo
