Setup

Choose the install path that matches your comfort level.

SafeClaw can be installed as a quick terminal command, a guided terminal flow, or a double-click Mac setup wizard. The global safeclaw command is opt-in so the installer does not silently modify a user’s shell environment.

Mac app after purchase

The easiest path for most Mac users is the packaged SafeClaw app from Gumroad.

The packaged DMG includes the SafeClaw runtime, so normal Gumroad users should not need Git, pip, Python setup, or Xcode Command Line Tools.

Get SafeClaw on Gumroad

1

Install the app

Download SafeClaw.dmg, open the DMG, drag SafeClaw into Applications, then open SafeClaw.

2

Choose a provider

Go to Setup and choose OpenAI, Ollama, Groq, OpenRouter / Claude, LiteLLM, or a custom OpenAI-compatible endpoint.

3

Add API settings

Enter the API key, base URL, and model name, then save config and run Doctor.

OPENAI_API_KEY=sk-your-key
OPENAI_BASE_URL=https://api.openai.com/v1
OPENAI_MODEL=gpt-4o-mini

If macOS keeps blocking the app after you move it into Applications, clear the download quarantine flag:

xattr -cr /Applications/SafeClaw.app

Quick install

This installs SafeClaw into ~/safeclaw, creates a virtual environment, installs dependencies, copies .env.example, and runs a local CLI check.

curl -fsSL https://raw.githubusercontent.com/amahmood561/SafeClaw/main/install.sh | bash

Guided terminal setup

Use this for non-dev users who want SafeClaw to ask for the install folder, model provider, workspace, permission profile, shell setting, and optional WhatsApp details.

bash <(curl -fsSL https://raw.githubusercontent.com/amahmood561/SafeClaw/main/guided-install.sh)

Mac double-click wizard

For a setup flow that feels closer to an app installer, clone or download the repo, open mac-setup, and double-click SafeClaw Setup.command. If macOS blocks it, right-click the file, choose Open, then confirm.

1

Pick install settings

The wizard asks where SafeClaw should live and whether to add the global command.

2

Choose model and safety defaults

Enter OpenAI-compatible settings, workspace path, permission profile, approval mode, and shell behavior.

3

Optional WhatsApp walkthrough

The wizard explains Twilio, public webhook URLs, sender allowlists, and persistent service mode.

macOS developer tools

If a source install, GitHub install, or setup script says xcode-select: note: No developer tools were found, install Apple Command Line Tools, then rerun the SafeClaw installer.

xcode-select --install

If the popup does not appear, reset the selected developer tools path and try again:

sudo xcode-select --reset
xcode-select --install

Use Ollama locally

SafeClaw can use Ollama through its OpenAI-compatible API. Start Ollama, pull a model, then point SafeClaw at the local base URL.

brew install ollama
ollama serve
ollama pull llama3.2:3b

OPENAI_API_KEY=ollama
OPENAI_BASE_URL=http://localhost:11434/v1
OPENAI_MODEL=llama3.2:3b

WhatsApp setup

WhatsApp mode uses Twilio. You need a Twilio Account SID, Auth Token, WhatsApp sender, public HTTPS URL, and allowed sender list. Run:

safeclaw whatsapp-setup https://your-public-url
safeclaw whatsapp --host 0.0.0.0 --port 8080

For persistent Mac mode, use the service commands after your webhook works locally.

safeclaw service-install
safeclaw service-start
safeclaw service-status

Verify your setup

Run doctor before debugging individual symptoms. It checks Python, environment config, model settings, workspace permissions, safety profile, Twilio config, port availability, and macOS service status.

safeclaw doctor
safeclaw run "write a one paragraph setup summary"
safeclaw chat