Open source solution to workos airlock. Currently in beta.
Why use this?
You don't want to inject real creds into your agents since most of the applications have really broad permissions. What waterlock gives you:
- A common language to work with permissions regardless of the application you choose.
- Full visibility into what the accesses were like / audit logs (upcoming).
- A gateway to proxy forward your external resource calls, preventing rogue agents from leaking your sensitive keys.
- Static TypeScript functions that can work for validation.
- Natural language for limiting what your agents can do. For example, with Gmail you can add the rule "No billing info allowed in emails", and a model you choose checks each call against it. These rules can only deny, never allow; see Checks.
- You own the integration logic. With WorkOS, you are coupled to their infra. This is a more flexible approach to integration.
- You own the AI layer: the AI SDK or Anthropic's SDK directly. Waterlock just works.
See it work
An AI SDK agent on a real Linear workspace, through the gateway. Creating the issue and setting its priority are allowed. Moving it to another team is blocked by a TypeScript rule. Filing a refund with a card number is blocked by a plain-language check. Linear is queried before and after the run.
Next
- Getting started: install, then a gateway in three steps.
- Concepts: Actions, Policies, Checks, Sessions, Credentials and the Request flow.
- Providers: Gmail and Linear.