Who it is forA developer who wants to contribute to Jan or needs a build tuned for their GPU.
The problemPrebuilt binaries may not match your hardware or let you change the code.
The outcomeA development build of Jan running locally with the engine variant you chose.
What you need first
- Windows 10+, macOS 13.6+ or most Linux distributions
- RAM for the model size: 8GB for 3B models, 16GB for 7B, 32GB for 13B on macOS
- To build from source: Node.js 20+, Yarn 4.5.3+, Make and Rust
Steps
- Install Node.js 20+, Yarn 4.5.3+, Make and Rust
- Clone with
git clone https://github.com/janhq/janandcd jan - Run
make devto install dependencies, build and launch the app - Pick an engine variant, for example
make dev JAN_ENGINE_VARIANT=cuda13 - On Windows, run make from Git Bash, not cmd.exe
Prompt for your agent
Build janhq/jan from source on this machine with the <engine variant> engine using make dev. If the build fails, read the error and apply the fix the README gives for that case.
Caveat. On Windows, CUDA builds can hit the 260 character path limit; the build script relocates the engine build tree, and otherwise the README says to set JAN_ENGINE_BUILD_DIR to a short path.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for janhq/jan and "Build Jan from source with your GPU engine"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer