Summary
Three installation methods. The simplest, and the one we recommend, is Docker: one file to download, one command, and Linkr is running — no repository to clone, nothing to compile. A manual install (Python + Node.js) gives the same thing without Docker, for development. Finally, client-only mode needs nothing but Node.js: more limited, but enough to discover Linkr or publish a static site.
Choose a method
Every feature, nothing to build. The images are published on Docker Hub: one file, one command, and Linkr is running. Also suits a server.
curl -O https://framagit.org/interhop/linkr/linkr/-/raw/main/docker/docker-compose.hub.yml
echo "LINKR_SECRET_KEY=$(python3 -c 'import secrets; print(secrets.token_urlsafe(48))')" > .env
docker compose -f docker-compose.hub.yml upThe same features, without Docker: the Python backend and the frontend run directly on your machine, with hot reload.
git clone https://framagit.org/interhop/linkr/linkr.git && cd linkr
npm install
python3 -m venv apps/api/.venv && source apps/api/.venv/bin/activate
pip install -e "apps/api[dev]"
# then the two configuration files, see the guide
npm run dev:allEverything runs in the browser: no backend, no accounts, no remote warehouse. To discover Linkr, work on local files on your own, or publish a static site.
git clone https://framagit.org/interhop/linkr/linkr.git && cd linkr
npm install
npm run dev:webComparing the three methods
| Docker | Manual | Client-only | |
|---|---|---|---|
| User accounts and sharing | ✓ | ✓ | — |
| Remote data warehouse | ✓ | ✓ | — |
| Git versioning | ✓ | ✓ | — |
| Nothing to compile | ✓ | — | — |
| Hot reload (development) | — | ✓ | ✓ |
| Publishing as a static site | — | — | ✓ |
What each mode enables is detailed in Deployment modes.
Just want to try it?
Nothing to install: the online demo runs Linkr right in your browser, with example data.
Connecting an AI agent
The Docker and manual installs also start Linkr’s MCP server, through which an agent acts on your projects: build a cohort, add a chart, suggest concept mappings. The agent runs in a chat client, installed separately.
We recommend LibreChat: open source, multi-user, and able to use only a model hosted in your institution. Install LibreChat gives its setup for your Linkr install. Client-only mode has no MCP server.
Next steps
- Have an agent work on Linkr: Install LibreChat.
- Deploy an instance for your team or institution: Production install.
- Build your first project: Your first project.
- Connect to a PostgreSQL, MySQL or Doris warehouse: Database connectors.