Welcome to SafeClaw
Your AI agent with built-in safety guardrails.
Every action gets checked before it runs. Your API keys and files never leave your machine.
Let's get you set up in 4 quick steps.
Your AI agent with built-in safety guardrails.
Every action gets checked before it runs. Your API keys and files never leave your machine.
Let's get you set up in 4 quick steps.
Both providers use the same safety model. Your API key never leaves your machine.
This key is saved securely on your machine and never shared with anyone — not even Authensor.
How to get your key:
Authensor is the safety layer that checks every action your AI agent takes. It only sees what the agent wants to do (like "write a file") — never your keys, passwords, or file contents.
Already have a token? Paste it above. Or request one manually.
Saving configuration and applying default policy...
SafeClaw is configured and ready. Your default deny-by-default policy is active.
Every action gets checked before it runs. Safe ones go. Risky ones need your OK. Dangerous ones are blocked.
Your keys and files never leave your machine. Only action descriptions are checked.
Runs the AI agent inside an isolated container (like a virtual machine) so it can only access files in one folder. Nothing outside that folder can be read, changed, or deleted — even if the agent tries.
When to use it: When running untrusted tasks, working with sensitive projects, or anytime you want an extra layer of protection.
What goes in the project folder: Everything the agent needs to do its job. Source code, config files, data files — if it's not in this folder, the agent can't see it. Your home directory, SSH keys, other projects, and system files are all completely off-limits. Think of it as handing someone a locked room with only the files you choose inside.
Setup: Requires Docker Desktop running on your machine. Install it, make sure it's open, and you're good to go.
Every action SafeClaw evaluates is permanently recorded here and cannot be altered after the fact.
Automate recurring tasks on a cron schedule with optional quiet hours.
Control what your AI agent can and cannot do. Rules are evaluated top-to-bottom.
See what would happen if your agent tried a specific action. Nothing actually runs.
Policy changes are automatically versioned. Roll back to any previous version.
Run a quick health check to make sure everything is working properly.