Open-source infrastructure for AI agents.
We research and build the open infrastructure for agents.
Products
PermissionAgentFGA
Fine-grained authorization for AI agents.
It decides what each agent can see and do, down to a single document, tool or record. Check every tool call, filter what reaches the model, and give agents access that expires. It speaks the OpenFGA API and runs on Cloudflare Workers, with one isolated store per customer.
agent.ts
import { OpenFgaClient } from "@openfga/sdk";
const fga = new OpenFgaClient({ apiUrl, storeId, credentials });
// Before the agent uses a tool, check that it may.
const { allowed } = await fga.check({
user: "agent:support-bot",
relation: "can_use",
object: "tool:issue-refund",
});
- Check latency at the edge
- 6msServer-side, when the store shares a data center with the request.
- Store size tested
- 1.1MRelationships in one store, with no measurable effect on check latency.
- Compared with OpenFGA
- 864,827Side-by-side requests against OpenFGA v1.22.0. Two differences, both caused by OpenFGA.
- Found along the way
- 9Bugs in OpenFGA itself, reproduced and written up.
Reports
Published reports.
Method, numbers and caveats, written so you can check them.
- Request-by-request compatibility with OpenFGA v1.22.0 864,827 random comparisons, 5,020 malformed and edge-case requests, OpenFGA's own conformance suite, and the 9 OpenFGA bugs the testing turned up.
- Performance on Cloudflare, and next to OpenFGA on one machine Check latency at 1.4k, 112k and 1.1M relationships per store, throughput per store and across stores, and a same-machine comparison with OpenFGA.
The front door is open.
We research and build the open infrastructure for agents. Read the code, run it yourself and check our work.