Run your AI agents on local models with BetterClaw Desktop
BetterClaw Desktop connects the models running on your computer, through Ollama, LM Studio, llama.cpp or any OpenAI-compatible address, to your BetterClaw agents in the cloud. Your agents keep working 24/7; your tokens are free. No open ports, no tunnels, no API key.
- Free to download
- Works with Ollama, LM Studio and llama.cpp
- Sign-in kept in your OS keychain
- Prompts never written to disk
All downloads
Pick your platform
Version 0.1.0, released 5 October 2026. Same app on every platform.
macOS
macOS 13 (Ventura) or later, Apple Silicon only (no Intel yet)
Windows
Windows 10 or 11, 64-bit, Intel or AMD (no Windows on ARM yet)
Linux
64-bit, Ubuntu 22.04+ or equivalent (AppImage only, no .deb or Flatpak yet)
What it does
What BetterClaw Desktop does
Your BetterClaw agents live in the cloud. They chat on Telegram, Slack, Discord and Gmail, run scheduled tasks and use 99 connectors, and normally they call an AI provider with your API key. BetterClaw Desktop gives them a second option: the models on your own computer. Install it, pick the models you want to share, and your agents use them instead of paying per token.
It finds your local models
BetterClaw Desktop detects Ollama, LM Studio and llama.cpp on your computer automatically, or connects to any custom OpenAI-compatible address on your machine or your local network. For each model it shows the context size and how well it handles agent tools, so you know before you assign it whether it can run a workflow or only chat.
You choose what your agents can use
Share one model or all of them. Set how many requests may run at once, and cancel any request from the app. Your agents then run on your computer's models; everything else about them stays the same.
It stays available in the background
The app starts when you log in, lives in your menu bar or system tray, and keeps your computer awake while it is answering an agent. Close the window and it keeps working. Quit it, or let the computer sleep or go offline, and your agents cannot reach your local models until it is back; agents on a cloud model are not affected.
You see activity, never content
The activity log shows each request's model, time taken and token count. It never shows the prompt or the answer, and it is cleared when you quit. Other apps on the same computer can use your local models through a private, key-protected API that is off by default.
No open ports
Why not just expose Ollama to the internet?
If you have tried to give a cloud agent access to a local model, you know the usual routes: set OLLAMA_HOST to 0.0.0.0 and open port 11434, run an ngrok or Cloudflare tunnel, or put the machine on Tailscale. They work, and they all have the same problem: your model server is now reachable from outside, usually with no authentication, and you maintain the tunnel, the firewall rule and the router forwarding yourself.
BetterClaw Desktop connects out to BetterClaw instead. You open no port, there is no public URL, and there is nothing to keep alive except the app itself.
| Feature | OLLAMA_HOST=0.0.0.0 + port forward | ngrok / Cloudflare Tunnel | Tailscale | BetterClaw Desktop |
|---|---|---|---|---|
| Inbound port on your machine | Yes (11434) | No, but a public URL | No (private network) | No |
| Authentication on the model API | None by default | Optional, you set it up | Network-level | Built in |
| Works from a cloud agent | Yes | Yes | Only if the agent is on your tailnet | Yes |
| Setup | Env var, firewall, router | Install, auth, keep the tunnel running | Install on both ends | Install, sign in, pick models |
| Shows tool support per model | No | No | No | Yes |
Last checked 8 October 2026.
Desktop or web
Desktop app or web app: which one do you need?
You need the web app to create and chat with your agents. Add BetterClaw Desktop when you want those agents to run on the models on your own computer.
| Feature | BetterClaw Desktop | Web app |
|---|---|---|
| What it is for | Connecting the AI models on your computer to your agents | Creating, running and chatting with your agents |
| Chat, agent setup, workflows, connectors, billing | No | Yes |
| Ollama, LM Studio, llama.cpp or a custom OpenAI-compatible address | Yes | Through BetterClaw Desktop |
| Cloud models with your own API key (Claude, GPT, Gemini, DeepSeek, 28+ providers) | Not needed | Yes |
| When you close the window | Keeps running in the menu bar or tray | Agents keep running in the cloud |
| When you quit, sleep or go offline | Agents cannot reach local models until it is back | Not affected |
| Price | Free to download | Plans on the pricing page |
Models
Which local models work for agents?
Any model your local server can run will chat. Agents need more: the model has to call tools reliably, follow a multi-step plan and keep a long context. BetterClaw Desktop flags tool support per model so you can see it before you assign one. As a rough guide:
- 7 to 9B models with tool calling (for example the current Qwen and Gemma small models) handle simple scheduled tasks and single-tool workflows. Expect to need at least 16 GB of RAM.
- 27B and up handle multi-step workflows with several tools. Plan on 24 GB of VRAM or unified memory, or accept slower runs.
- Models without tool calling can still be used for chat-only agents.
For tested picks and the exact models we would skip, read Best free models for OpenClaw agents and which Ollama models support tool calling. The same results apply here.
Requirements
System requirements
The BetterClaw desktop app runs on macOS 13 (Ventura) or later on Apple Silicon, 64-bit Windows 10 and 11 on Intel or AMD, and 64-bit Linux as an AppImage. On Windows it is a regular app: no WSL, Docker, Virtual Machine Platform or virtual machine needed. The only other thing you need is a local model app such as Ollama or LM Studio.
| Platform | Requirements | File | Size |
|---|---|---|---|
| macOS | macOS 13 (Ventura) or later, Apple Silicon only (no Intel yet) | DMG | 128 MB |
| Windows | Windows 10 or 11, 64-bit, Intel or AMD (no Windows on ARM yet) | EXE | 111 MB |
| Linux | 64-bit, Ubuntu 22.04+ or equivalent (AppImage only, no .deb or Flatpak yet) | AppImage | 128 MB |
| Intel Mac | Not supported. Use the web app with a cloud model instead. | ||
| Memory and disk | The app itself is light. What you need depends on the models you run: as a rough guide, 7 to 8B models want at least 16 GB of RAM. | ||
Getting started
Install in under a minute
Download the file for your platform and run it. If BetterClaw Desktop doesn't find Ollama, LM Studio or llama.cpp, it links you to a download page.
macOS
- 1Open BetterClaw.dmg.
- 2Drag BetterClaw into Applications.
- 3Launch it from Applications or Spotlight. The app is signed and notarized by Apple, so it opens with a double-click.
Windows
- 1Run BetterClaw.exe.
- 2If Windows shows "Windows protected your PC", click More info, then Run anyway.
- 3Follow the installer prompts.
- 4Launch BetterClaw from the Start menu.
Linux
- 1Make it executable: chmod +x BetterClaw.AppImage
- 2Run it: ./BetterClaw.AppImage
- 3If it fails with a FUSE error, install libfuse2 (sudo apt install libfuse2, or libfuse2t64 on Ubuntu 24.04).
- 4To add it to your app menu, open it with AppImageLauncher or Gear Lever.
Then connect an agent
- 1Sign in with your BetterClaw account.
- 2Tick the models you want to share.
- 3In the web app, open an agent's model setting and pick one of your local models.
- 4Run a task. The request shows up in the activity log on your computer.
Privacy
Where your sign-in and data are stored
BetterClaw Desktop keeps your sign-in in your computer's built-in secure storage: Keychain on Mac, Credential Manager on Windows and Secret Service on Linux. It never asks for an AI provider API key, because your models run on your own computer.
There is no local database. App settings, such as which models you share, are saved in a preferences file at ~/Library/Application Support/BetterClaw Desktop/ on Mac, %APPDATA%\BetterClaw Desktop\ on Windows and ~/.config/BetterClaw Desktop/ on Linux. This file is not encrypted. Keys you create for the local API are saved there as one-way hashes, never as the keys themselves, and never leave your machine.
Prompts and answers are never written to disk, and the activity log is cleared when you quit. When an agent uses one of your local models, its request reaches your computer through BetterClaw, and BetterClaw does not store the content of prompts or answers.
Desktop app questions
Is the BetterClaw desktop app free to download?
Yes. The installers for macOS, Windows and Linux are free to download. BetterClaw plans are listed on the pricing page.
Can I run BetterClaw agents on Ollama?
Yes. Install Ollama and BetterClaw Desktop on the same computer, tick the models you want to share, and pick one of them in the agent's model setting. No OLLAMA_HOST changes, no port forwarding.
Does it work with LM Studio or llama.cpp?
Yes. The app detects LM Studio and llama.cpp servers automatically, and you can add any OpenAI-compatible address on your computer or local network.
Do I need to expose Ollama to the internet or set up a tunnel?
No. BetterClaw Desktop connects out to BetterClaw, so you do not open a port, forward your router, change OLLAMA_HOST or run a tunnel.
Which local models support tool calling for agents?
The app shows tool support for each model it finds. As a rule, current small Qwen and Gemma models with tool calling handle simple tasks; 27B and larger models handle multi-step workflows. See our tested list of free and local models for agents.
Does the BetterClaw desktop app work on Intel Macs?
No. The macOS build is for Apple Silicon only (M1 and later) and needs macOS 13 (Ventura) or later. An Intel Mac version is not available yet. On an Intel Mac, use the web app with a cloud model.
Is there a Linux version?
Yes. BetterClaw ships a 64-bit AppImage for Linux. .deb and Flatpak packages are not available yet.
Do my agents stop when I close my laptop?
Agents using a local model pause until the app is running again; agents using a cloud model keep going. The app keeps your computer awake while it is answering a request, but it cannot stop a closed laptop from sleeping.
Where does BetterClaw Desktop store my sign-in and data?
Your sign-in is kept in your computer's secure storage: Keychain on Mac, Credential Manager on Windows and Secret Service on Linux. There is no local database. Settings are saved in a preferences file, prompts and answers are never written to disk, and the activity log is cleared when you quit.
How do I update or uninstall it?
On Mac, BetterClaw Desktop checks for updates in the background and installs them the next time you quit the app. On Windows and Linux, download the latest version from this page. To uninstall, drag it to the Trash on Mac, remove it under Settings > Apps > Installed apps on Windows, or delete the AppImage on Linux. To remove its settings too, delete ~/Library/Application Support/BetterClaw Desktop/ on Mac, %APPDATA%\BetterClaw Desktop\ on Windows, or ~/.config/BetterClaw Desktop/ on Linux.
Rather not install anything?
BetterClaw also runs entirely in the browser with cloud models and your own API key, and your agents stay on 24/7 either way. Open the web app, or see the plans on the pricing page.