SafeClaw by AceTeam

Turn safety on for your AI agents. Real-time monitoring and enforcement. Runs in a sandboxed container on your computer. Can't access your files, email, or passwords.

Safety Off vs. Safety On

Safety Off
  • No spending visibility
  • Agent can delete files, scan ports, run exploits
  • Personal data can leak to any API
  • No record of what happened
Safety On
  • See cost per call in real time
  • Dangerous actions blocked before they happen
  • SSNs, credit cards, and PII caught and stopped
  • Every decision signed and recorded

SafeClaw watches for: data leaks, unauthorized access, dangerous commands, cost spikes, and prompt injection attempts.

OpenClaw NemoClaw NanoClaw CrewAI LangChain Claude Code Any LLM API

Setup (5 minutes)

Step 1 — Install a container runtime

SafeClaw runs in an isolated container on your computer. Install one of these:

Podman (recommended) — open source, rootless by default: podman.io/docs/installation

Docker Desktop — also works: docker.com/products/docker-desktop

Both are available for Mac, Windows, and Linux. Podman runs without root access, which means stronger isolation. Either one takes about 2 minutes to install.

Step 2 — Open a terminal

Mac: Open the app called Terminal (search for it in Spotlight with Cmd+Space).

Windows: Open PowerShell (search for it in the Start menu).

Linux: Open your terminal emulator.

Step 3 — Start SafeClaw

Paste this command and press Enter:

podman run -p 8899:8899 ghcr.io/aceteam-ai/aep-proxy

Using Docker? Replace podman with docker — the command is identical. First run downloads about 1 GB. After that, it starts in seconds.

Step 4 — Open the dashboard

Go to localhost:8899/aep in your browser.

The setup wizard will guide you through configuring your API key. After that, you'll see cost tracking, safety signals, and policy controls in real time.

Step 5 — Point your agent at the proxy

Add this to your terminal so your agent's LLM calls go through SafeClaw:

export OPENAI_BASE_URL=http://localhost:8899/v1

Every LLM call from any tool (OpenClaw, CrewAI, LangChain, custom scripts) now goes through SafeClaw automatically.

In a workshop? Your instructor will give you a proxy URL. No install needed — just point your agent to the URL and open the dashboard in your browser.

Example:
export OPENAI_BASE_URL=http://INSTRUCTOR_URL:8899/v1

Replace INSTRUCTOR_URL with the address your instructor provides.

Claude Code Integration

SafeClaw provides safety tools directly inside Claude Code via MCP. Add this to your Claude Code config:

{"mcpServers":{"aceteam":{"type":"streamable-http","url":"http://localhost:8899/mcp/"}}}

Claude Code gets 4 safety tools: check_safety (scan text for threats), get_safety_status (session metrics), set_safety_policy (toggle detectors), and get_cost_summary (spending breakdown).

Quick setup: If you have Python 3.12+, you can skip the manual config:
pip install aceteam-aep[all] && aceteam-aep setup

This detects your container runtime, starts the proxy, configures Claude Code, and opens the dashboard — all in one command.

What Does SafeClaw Do?

SafeClaw ships with 5 detection categories: Finance, IoT, Software, Web, and Program. Each can be toggled independently. Enterprise customers get custom categories and calibrated detection models.

Is This Free?

Yes. SafeClaw is open source (Apache 2.0). You can read every line of code on GitHub. The proxy, safety detectors, dashboard, and signed audit trail are all free.

We offer paid services for companies that need managed hosting, custom safety policies, and compliance reporting. The open source version is the full product, not a limited trial.

Want to try AceTeam? Free accounts get $5 of LLM credit and access to 40+ workflow node types. Connect from the dashboard or run aceteam-aep connect.

Troubleshooting

"podman" or "docker" is not recognized / command not found
Your container runtime isn't installed or isn't running. Install Podman (recommended) or Docker Desktop. If using Docker, make sure the app is open (whale icon in your taskbar).
The download is very slow
The first download is about 1 GB. This is a one-time thing — after that, SafeClaw starts in seconds. If you're on slow WiFi, try a wired connection or hotspot.
I can't open localhost:8899
Make sure the podman run (or docker run) command is still running in your terminal. If it stopped, run it again. Also check that port 8899 isn't used by another app.
I want to use pip instead of containers
If you have Python 3.12+, you can run SafeClaw directly on your machine:

pip install aceteam-aep[all] && aceteam-aep proxy --port 8899

Note: this runs on your host, not in a container. The proxy will have access to your filesystem. Podman or Docker is recommended for isolation.

Host a Workshop

Anyone can run a SafeClaw workshop. Here's what you need:

Instructor Setup

1. Install and start the proxy on your machine (the one attendees will connect to):

pip install aceteam-aep[all]
export OPENAI_API_KEY=sk-your-key
aceteam-aep proxy --port 8899 --host 0.0.0.0

2. Share your URL with attendees. On the same network (WiFi):

export OPENAI_BASE_URL=http://YOUR_IP:8899/v1

For remote attendees, use ngrok:

ngrok http 8899

3. Open the dashboard at http://YOUR_IP:8899/aep/ — project it on screen so everyone sees calls flowing in real-time.

Attendee Setup

Attendees only need one command:

export OPENAI_BASE_URL=http://INSTRUCTOR_URL:8899/v1

Then use any tool — SafeClaw, OpenClaw, Python, curl — all LLM calls route through the safety proxy automatically.

Demo Flow (30 min)

5 min — Connect to proxy, see the dashboard, make a normal call (PASS)
5 min — Send a dangerous request → BLOCKED. Toggle safety off → same request passes. Toggle on → blocked again.
10 min — Explore 5 safety categories: Finance, IoT, Software, Web, Program
10 min — Set custom policies, discuss enterprise use cases

Full workshop materials (slides, scripts, bootcamp guide)
Workshop materials — slides, speaker script, bootcamp guide, demo scripts.
hosted-proxy.sh — automated setup script for instructors.

Learn More

GitHub AEP Protocol PyPI AceTeam
SafeClaw by AceTeam · Open source (Apache 2.0)