Runtime Agent Governance for AI Agents in JavaScript and Node.js
Cerone launched agent-governance, an npm package that provides runtime governance for AI agents in JavaScript and Node.js. The package communicates with the Cerone runtime to return explicit decisions—approved, flagged, or rejected—before an agent action executes. Developers can start immediately with hosted trial access from the SDK, validate agent actions in real time, and keep their own OpenAI, Anthropic, or other model keys. The tool adds runtime governance without replacing the rest of the stack, offering a lean trust layer instead of a heavy platform rewrite. Installation requires Node 18+ due to reliance on built-in fetch and AbortController. Usage involves creating an agent with a description and allowed actions, then validating specific actions with context. For example, a customer billing support agent with permissions for database reads and billing API calls can be validated for a database query on a billing table. The package aims to provide real decisions instead of vague policy claims.
Developers can add runtime safety to AI agents without overhauling their stack.