[ docs ] // Reference

Commands

Everything ArcFlare understands, in one place. arcflare help prints the short version.

Models and chat#

Command	Does
arcflare	The menu: harness → model → context → auto mode
arcflare ls	List discovered GGUF models (aliases: list, models)
arcflare pull <user/repo>[:QUANT]	Download a GGUF from Hugging Face
arcflare run <model> [--rc]	Load it and chat. arcflare <model> is the same
arcflare fit [model] [--run]	What fits; with --run, measure what runs best

Command	Does
arcflare agent [model]	The coding agent. -p "…" one-shot, --yolo/--ask, --rc, --machine/--no-machine
arcflare use <harness> [model]	Configure and launch opencode, hermes, hermes-desktop, codex, chat or agent
arcflare use … --no-launch	Write the config only
arcflare mcp	The machine server on stdio
arcflare mcp --install [--project]	Register it in ~/.arcflare/mcp.json (or this folder)
arcflare mcp enable \| disable	Install the machine server, or leave it out
arcflare mcp trust \| untrust \| trusted	Allow this folder's .mcp.json to start servers
arcflare mcp login \| logout [server]	Sign in to a hosted MCP server

Command	Does
arcflare gen	Every generator and whether it's ready
arcflare gen setup <model>	--torch cuda\|cpu\|rocm, --torch-from <python>, --texture
arcflare gen 3d <image>	Image → .glb. --prompt, -m, -o, --steps, --octree, --faces, --seed, --texture
arcflare gen tts "text"	Text → .wav. -m, -o, --voice, --lang, --instruct, --ref, --ref-text, --file
arcflare gen check <model>	Can this machine run it, and if not, why
arcflare gen python [path]	Show, or set, the Python that runs generators
arcflare gen default <model>	The model gen 3d uses without -m

Command	Does
arcflare shop [search]	Browse the model hub: what fits your GPU, what to get (--fits, --cat, --json, show <model>)
arcflare uninstall	Remove ArcFlare and its files; shows everything first (--dry-run, --models, --keep-cli)
arcflare rc	Your remote-control key, relay and link
arcflare rc relay [url]	Show or set the relay
arcflare rc rotate	A new key; every paired device is cut off
arcflare update	Install the latest version
arcflare update --check	Check now, install nothing
arcflare update --pack [dir]	Make an offline update tarball
arcflare update --from <tgz\|dir>	Install from a tarball or folder, with no network
arcflare update --off \| --on	Background update checks

Command	Does
arcflare serve [--port N] [--restart]	Start the OpenAI-compatible server only (default port 11434)
arcflare ps	Is it running, and what is loaded
arcflare stop	Stop it
arcflare logs [-n N]	Tail the server log
arcflare backend [kind] [--probe]	List or select a llama.cpp backend
arcflare memory [lean\|balanced\|max]	Prompt cache, resident models, idle sleep
arcflare batch [512\|1024\|2048\|4096]	Physical batch size
arcflare engine install [--vulkan\|--cuda\|--cpu\|--rocm] [--if-missing]	Download llama.cpp for this machine; --if-missing leaves one you have alone
arcflare set-engine <path>	Remember where llama-server lives
arcflare doctor	Check engine, GPU, models, harnesses, Blender
arcflare path · version · help	Home directory, version, help

Type	In	Does
/rc [off\|new]	chat, agent	Remote control
/update [tgz\|dir]	chat, agent	Install an update (or from an offline source)
/stats	agent	Steps, tool calls, tok/s, prompt size
/tools	agent	Built-in and MCP tool counts
/skills	agent	Discovered skills
/help	chat, agent	This list
/bye /exit /quit	chat, agent	Leave

Variable	Does
ARCFLARE_HOME	Where ArcFlare keeps its files (default ~/.arcflare)
ARCFLARE_MODELS	An extra folder of GGUFs, searched first
LLAMA_CACHE	llama.cpp's download cache; also where pull saves
ARCFLARE_LLAMA_SERVER	A specific llama-server binary
ARCFLARE_LLAMA_DIR	An extra folder to search for llama.cpp builds
ARCFLARE_NO_MACHINE=1	Don't connect the machine server to the agent
ARCFLARE_MCP_ROOTS	Folders the machine server may run commands in
ARCFLARE_BLENDER / ARCFLARE_BLENDER_PORT	Which Blender, and its bridge port
ARCFLARE_PYTHON	The Python that runs 3D generators
HY3DGEN_MODELS / HF_HUB_CACHE	Where generator weights are kept
ARCFLARE_RELAY	The remote-control relay, for this run
ARCFLARE_NO_UPDATE_CHECK=1	No background update checks
ARCFLARE_DEBUG=1	Save the last agent request for inspection
NO_COLOR	Plain output

Everything lives in ~/.arcflare unless ARCFLARE_HOME says otherwise.

File	Holds
config.json	Your choices: last model and harness, auto mode, memory, batch, machine server
models.ini	Router presets: context and cache type per model
fit.json	Measurements from fit --run
backends.json	The last backend probe
mcp.json	Your MCP servers
trust.json	Which folder MCP configs you approved, and their fingerprints
oauth.json	Tokens for hosted MCP servers
rc.json	Your remote-control key and relay (keep it private)
update.json	The last update check
skills/	Your skills
gen/	3D generation: venv, model code, weights, output
server.log	llama-server's log

// One command not listed above: type arcflare on any page of this site (not in a text box). No terminal needed.