I’ve been experimenting with AI agents lately, and one problem kept coming up: they either get a raw API key with full access or nothing at all. That’s risky, especially if you’re testing agents that can make arbitrary calls.
So I hacked together a tiny package called Kage Keys - https://github.com/kagehq/keys
It lets you wrap agent actions with scoped, short-lived tokens instead of handing over your real API keys.
Example:
```js
import { withAgentKey, getLogs } from "@kagehq/keys";
async function main() {
await withAgentKey("github:repos.read", async () => {
console.log("Agent is calling GitHub API...");
});
console.log(await getLogs());
}
main();
Right now it:
- Generates scoped, expiring tokens (default 10s)
- Logs every action to kage-keys.log
- Works as a drop-in wrapper for async functions
It’s just an MVP (tokens are fake UUIDs), but I want to see if developers find this helpful before building the production version with real crypto + proxy enforcement.
Repo: https://github.com/kagehq/keys
npm: https://www.npmjs.com/package/@kagehq/keys
Would love feedback, especially from anyone running agents in production or dealing with API key sprawl.
