Companies are racing to deploy agents that don't just answer questions, they take actions: calling APIs, moving money, changing records. The tooling today is all observability; you find out what the agent did after it already did it. That's a camera pointed at the vault, not a lock on the door.

Verge sits at the function call boundary and intercepts every tool call before it executes, evaluates it against a no-code policy ruleset, and either allows it, blocks it, or freezes it pending human approval, right in Slack or a web dashboard, with a reasoning trace so the approver can see why the agent wanted to act. It plugs into LangGraph, LangChain, the Claude API, and the OpenAI API, and stays a pure governance layer across all of them.

The core demo works end to end: an agent tries a tool call, Verge catches it, freezes it, routes it for approval, and resumes once it clears. Current work is packaging the SDK and building the no-code policy builder.