Local models

BetterClaw Desktop connects your computer to BetterClaw as a device and shares models from a local AI engine — Ollama, LM Studio, or llama.cpp — with your agents. The agent keeps running in BetterClaw as usual; only its model calls are answered by your computer.

The app dials out to BetterClaw — it never opens a port to the internet, so there are no tunnels, public addresses, or API keys to set up.

What you need

• BetterClaw Desktop on the computer that will serve the model. It's currently built for Macs with Apple silicon. Find the download from the Download BetterClaw Desktop link in the Local LLM Models card on an agent's Model tab.
• A local engine, installed and running: Ollama, LM Studio (with its local server turned on), or llama.cpp (llama-server). Any other OpenAI-compatible server on the same computer also works as a custom engine address. BetterClaw Desktop finds engines but never installs or updates them.
• At least one model downloaded in that engine.
• The computer must be awake, online, and running BetterClaw Desktop whenever the agent works.

Connect your computer

1. Open BetterClaw Desktop and click Sign in with BetterClaw.
2. Your browser opens the Connect BetterClaw Desktop page. Check the computer's name and click Connect this computer.
3. The app picks up the sign-in on its own. If it doesn't, copy the code shown on the page, paste it into the app, and click Continue — the code works once and expires in two minutes.

Once connected, the app's sidebar shows Connected, and the computer appears under Your computers on your profile (open the menu under your name at the bottom of the left rail and choose Your profile).

Choose an engine and share models

In BetterClaw Desktop:

1. Open Engines. Each engine shows Running, Installed · not running, or Not installed (with a Download button). Start your engine and click Check again — no restart needed. If more than one is running, click Use on the one you want. For another OpenAI-compatible server, open Advanced: custom engine address and enter its address (for example http://127.0.0.1:8000).
2. Open Models and turn on the models your agents may use. Only models you share are visible to BetterClaw. Each model shows its context window and how well it's likely to handle tools — Good at tools, Weak at tools, or Tool use unknown.
3. Make sure Local Inference is on under Capabilities (it is by default).

Use a local model for an agent

1. Open Settings ▸ Agents, select the agent, and open its Model tab.
2. In the Local LLM Models card, open Local model and pick a computer and model — each entry shows whether the computer is Online or Offline, its context size, and a warning if the model may be weak at tool use.
3. Click Use this model. The agent restarts and runs on the model on your computer.

The card then shows the computer and model as Active. On a paid plan the local model also appears as the Primary in your connected models, labelled Local model.

To go back to a cloud model, choose or connect one in the model settings above the card — a local model isn't kept as a backup when you switch away from it.

Who can use it

A computer belongs to the person who signed in on it. Only you can point an agent at your own computer, and you only see your own computers in the picker — teammates can't use yours. If an agent is on a computer that was removed or belongs to a teammate, the card says so and asks you to choose another model.

On workspaces that require you to bring your own LLM, a local model counts as your own.

When your computer is offline

• With backup models (paid plans) — the agent switches to its backup models while the computer is offline. Add one under connected models.
• Without a backup — chats with the agent fail until the computer reconnects.

When a local model can't answer, the chat shows why — for example "The computer serving this local model is offline. Open BetterClaw Desktop on it to reconnect." or "The computer serving this local model is busy with another request."

Settings in BetterClaw Desktop

• Requests at once (under Capabilities) — how many requests the computer handles in parallel, from 1 (the default) to 8. Extra requests get a "busy" answer.
• Keep awake while serving — stops the computer from sleeping while a request is running (on by default).
• Open at login — starts the app in the menu bar when you log in.
• Device name — how the computer appears in BetterClaw.
• Activity — a list of events (sign-ins, connections, requests started, completed, or refused). What was asked and answered is never shown.
• Local API (advanced) — off by default. Turn it on to use your shared models from other OpenAI-compatible tools on the same computer, at http://127.0.0.1:<port>/v1. Every request needs a key you create there; a new key is shown only once, and you can revoke it anytime. It's reachable only from that computer.

Closing the window leaves the app running in the menu bar; your agents can use the computer until you quit the app.

Manage or remove a computer

Under Your computers on your profile, each computer shows whether it's online, its app version, when it was last seen, and the models it shares. From there you can:

• Activity — recent requests with model, result, tokens, and time.
• Rename the computer.
• Remove it from your account.

Signing out in BetterClaw Desktop (Settings ▸ Sign out) also removes the computer. Agents using its models switch to their backup models.

Troubleshooting