CLI Reference


Usage: jan <COMMAND>
Commands:
  serve    Load a local model and expose it at localhost:6767/v1
  launch   Start a local model, then launch an AI agent with it pre-wired
  threads  List and inspect conversation threads
  models   List and load models installed in the Jan data folder
  agent    Run the autonomous agent with tool use

`jan serve`

Load a local model and expose it at localhost:6767/v1 as an OpenAI-compatible API.


jan serve [MODEL_ID] [OPTIONS]

Option	Description	Default
`MODEL_ID`	Model ID or HuggingFace repo (e.g. `unsloth/Qwen3-4B-GGUF`)	interactive picker
`--model-path`	Direct path to model file	—
`--bin`	Path to inference binary	auto-discovered
`--port`	Port to listen on	`6767`
`--mmproj`	Path to multimodal projector	—
`--embedding`	Treat as embedding model	`false`
`--timeout`	Request timeout in seconds	`120`
`--n-gpu-layers`	GPU layers to offload (-1 = all, 0 = CPU)	`-1`
`--ctx-size`	Context window size in tokens	`32768`
`--fit`	Auto-fit context to available VRAM	`false`
`--threads`	CPU threads (0 = auto)	`0`
`--api-key`	API key required by clients	`""`
`-d, --detach`	Run in background, print PID	`false`
`--log`	Log file path (for detached mode)	—
`-v, --verbose`	Print full server logs	`false`
`--select`	Show quantization picker when downloading	`false`

Examples:


jan serve                              # pick a model interactively
jan serve qwen3-4b                     # serve a specific model
jan serve unsloth/Qwen3-4B-GGUF       # auto-download from HuggingFace
jan serve qwen3-4b --fit              # auto-fit context to VRAM
jan serve qwen3-4b --detach           # run in background
jan serve qwen3-4b --port 8080        # custom port

When given a HuggingFace repo ID (e.g. owner/repo), Jan Agent automatically downloads the GGUF model if it isn't already installed.

`jan launch`

Start a local model server, then launch an external AI agent with environment variables pre-configured.


jan launch [PROGRAM] [OPTIONS]

Option	Description	Default
`PROGRAM`	Agent to launch: `claude`, `openclaw`, `opencode`	interactive picker
`--model`	Model ID to load	interactive picker
`--bin`	Path to inference binary	auto-discovered
`--port`	Port for the model server	`6767`
`--api-key`	API key (exported as `OPENAI_API_KEY`)	`jan`
`--n-gpu-layers`	GPU layers to offload	`-1`
`--ctx-size`	Context window size	—
`--fit`	Auto-fit context to VRAM	auto for `claude`
`-v, --verbose`	Print full server logs	`false`
`--select`	Show quantization picker	`false`

Examples:


jan launch claude                       # pick a model, launch Claude Code
jan launch claude --model qwen3-4b     # use specific model with Claude Code
jan launch openclaw --model qwen3-4b   # wire OpenClaw to a local model
jan launch opencode --model qwen3-4b   # wire opencode to a local model

The launch command sets these environment variables for the spawned program:

Variable	Value
`OPENAI_API_KEY`	The `--api-key` value
`OPENAI_BASE_URL`	`http://localhost:<port>/v1`
`ANTHROPIC_AUTH_TOKEN`	The `--api-key` value
`ANTHROPIC_BASE_URL`	`http://localhost:<port>/v1`

`jan models`

List and manage models installed in the Jan data folder.


jan models <SUBCOMMAND>

`jan models list`


jan models list [--engine <ENGINE>]

Option	Description	Default
`--engine`	Filter by engine	`all`

`jan models load`

Load and serve a model (alias for jan serve).


jan models load [MODEL_ID] [OPTIONS]

Accepts the same options as jan serve.

`jan threads`

List and inspect conversation threads saved by the Jan app.


jan threads <SUBCOMMAND>

Subcommand	Description
`list`	Print all threads as JSON
`get <ID>`	Print a thread's metadata as JSON
`messages <THREAD_ID>`	Print all messages in a thread as JSON
`delete <ID>`	Permanently delete a thread and its messages

`jan agent chat`

Run the built-in autonomous ReAct agent.


jan agent chat [OPTIONS]

Option	Description	Default
`--api-url`	LLM endpoint URL	auto-serve local model
`-m, --message`	One-shot prompt (skip interactive TUI)	—
`--mount`	Directories to expose to the agent (repeatable)	—
`--agent`	Agent implementation to use	`react`
`--sensor-url`	Sensor input endpoint URL	—

Plus all jan serve options (model ID, port, ctx-size, etc.) when auto-serving.

Examples:


# Interactive TUI with external endpoint
jan agent chat --api-url http://localhost:6767/v1
# One-shot with auto-served model
jan agent chat --model qwen3-4b -m "List all TODO comments in this project"
# Mount specific directories
jan agent chat --api-url http://localhost:6767/v1 --mount ./src --mount ./tests

Error Strategy Configuration