Quickstart
Get an AI agent running on your local hardware in three commands.
Build Jan Agent
git clone https://github.com/janhq/jan-agent.gitcd jan-agentcargo build --release -p jan-cli
Serve a model
If you have models installed via Jan Desktop:
./target/release/jan serve
An interactive picker shows your installed models. Select one and it starts serving at localhost:6767/v1.
If you don't have any models yet, provide a HuggingFace repo ID to auto-download:
./target/release/jan serve unsloth/Qwen3-4B-GGUF
Launch an agent
Wire Claude Code (or another agent) to your local model:
./target/release/jan launch claude
Jan automatically starts a model server and launches Claude Code with the correct environment variables pre-configured.
One-shot agent mode
Run the built-in ReAct agent for a single task:
./target/release/jan agent chat --api-url http://localhost:6767/v1 -m "List all Rust files in this project"
Or start an interactive agent TUI session:
./target/release/jan agent chat --api-url http://localhost:6767/v1
When using agent chat without --api-url, Jan Agent will automatically start a local model server for you.
What's next?
- CLI Reference — Explore all available commands and options
- Architecture — Understand how the crates work together
- Configuration — Customize agent behavior with config files