What it does
Jev steers rather than authorizes: a repeated write can pass its one-time hold and still needs the app’s approval. Failed checks continue without a verdict, and spending is counted after responses rather than strictly capped.
Agent tooling
@ghostmind-dev/agento
Build a tool-using assistant with a writing model and Jev checkpoints that suggest when to reconsider a result, change or answer.
Bookmark: @ghostmind-dev/agento
Keep this in your collection.
Leave a noteWhat would you try with this?
: @ghostmind-dev/agento
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
Have a developer start with the example that reads information and asks permission before changing it. The default answers are scripted, not paid AI. For your own assistant, keep that approval step separate from Jev’s suggestions. Your app decides which tools the assistant has and who may approve changes.
Live use needs a key for OpenRouter, an outside AI service. The writing model sees conversations and tool results. Jev checks shortened passages, but the assistant can also send it information of its own choosing. Repeating a proposal can pass Jev’s one-time hold. Failed checks and spending settings do not guarantee safety.