Installation
Requirements, ways to install, platform support, and running without network access.
Requirements
| Requirement | |
|---|---|
| Node.js | 22.13 or newer |
| Operating system | macOS (Apple Silicon and Intel), Linux (x64 and arm64), Windows (x64) |
| Disk | About 750 MB for the npm cache on first install, 23 MB for the embedding model, and an index of roughly 7 KB per chunk of code |
| Network | Only to download the package and, once, the embedding model |
No compiler is needed on the supported platforms: the SQLite engine, the vector extension, and the parsers all ship prebuilt.
Ways to install
Nothing to install up front. npx downloads remimem on first use and caches it:
npx -y remimem init
npx -y remimem serveMCP client configurations use the same command, so the server always matches the version npx resolves.
Check the installation
npx -y remimem doctorA healthy machine prints:
Node.js: 22.x.x ok
Grammar javascript: ok
Grammar typescript: ok
Grammar tsx: ok
Grammar python: ok
Grammar go: ok
SQLite with FTS5 and sqlite-vec: ok
Embedding model: local:Xenova/all-MiniLM-L6-v2 ok (384 dimensions)
Index: none yet (run `remimem init` in the project)Any failed check prints failed (...) with a fix hint, and doctor exits with code 1. In an indexed project, the last line shows files, chunks, pending changes, and chunks still waiting for a vector.
The embedding model
The first init, index, or doctor downloads Xenova/all-MiniLM-L6-v2 (23 MB) from huggingface.co into ~/.cache/remimem/models. After that, remimem needs no network.
- To keep the model somewhere else, set
REMIMEM_CACHE_DIR. The model goes into itsmodelssubfolder. - To work without any model, set
"embedding": { "provider": "none" }in.remimem/config.json. Search then uses keywords and exact symbol names only, which is still useful, but less so for questions phrased in plain language. - To use a hosted model instead, see Embedding providers.
Offline and air-gapped machines
On a machine with network access, run npx -y remimem doctor once so the model is downloaded.
Copy ~/.cache/remimem/models to the same path on the offline machine, or to a folder you point REMIMEM_CACHE_DIR at.
Install remimem there through your internal npm registry or mirror. Its dependencies (the SQLite engine, the vector extension, and the model runtime) are installed with it, so a lone package tarball is not enough. Then run remimem doctor to confirm the model loads.
Uninstall
remimem keeps its data in three places. Remove any of them safely:
| Path | What it is |
|---|---|
.remimem/ in each project | Config, index, and memories. Keep memory.json if your team shares it. |
~/.cache/remimem/ | The embedding model and benchmark clones. |
| Your MCP client configuration | Remove the remimem entry, for example claude mcp remove remimem. |
A global install is removed with npm uninstall -g remimem.