OpenVerb
The universal execution layer for AI agents
Start new thread

OpenVerb - Define AI actions as executable verb schemas

by•
OpenVerb lets developers define structured “verb schemas” that describe what AI agents are allowed to do. Instead of relying on prompt guessing, AI actions are validated, policy-checked, and executed through a deterministic runtime. Build libraries of workflows, test them in a live execution sandbox, and safely connect AI to real systems.

Add a comment

Replies

Best
Maker
📌
We built OpenVerb because AI agents became powerful, but unpredictable. Prompts alone aren’t enough when systems need to execute real actions safely. OpenVerb introduces a simple idea: define what AI can do through structured verb schemas. Each action is: • validated against a schema • checked against policies • executed through a deterministic runtime • traceable via a full execution log To make this understandable, we also built a live execution sandbox where you can: • paste a verb library • send a prompt • and watch the full execution pipeline in real time Would love feedback from builders working on AI systems, agents, or backend automation.

the schema-first approach feels really thoughtful, mapping natural language straight into typed verbs instead of vibes feels like the right call for an open-source execution layer.

Maker
Thanks! I completely agree. I think the biggest challenge now is adoption, most AI systems are still using the conventional prompt only approach. Hopefully OpenVerb can help move the ecosystem toward typed, reliable execution.

Dropping an upvote, congrats big shoutout for building a live sandbox right at launch, makes it so much easier to understand the runtime logic.

Maker
Thanks, I did recently add the sandbox and the new libraries page where people can publish verb libraries they think are interesting or verbs that they have worked with personally.

The verb definition setup clicked faster than I expected, and seeing an agent trigger a real action in my test app was a nice moment. Wish the docs had a couple more end-to-end examples though.

How does OpenVerb actually prevent an AI agent from misusing exposed verbs, like is there built-in auth scoping or do we have to wire that up ourselves on our end?

Maker

 Great question, yes you can write that up yourself. The great thing about openverb is that it gives you all the control. You define all your actions explicitly then you can build polocies around that such as paywall tiers, auth tiers etc.