Omnafy documentation

How to wrap your systems in a governed tool catalog, give agents an identity, and run them under supervision you can prove.

Getting started

If you are new to the gateway, work through these three in order. Together they take about an afternoon.

The five things the gateway manages: identities, tools, risk tiers, runs, and approvals. Read this first; everything else assumes it.

Wrap an existing API endpoint as a typed operation, declare its inputs, assign it a risk tier, and call it from an agent.

Write an agent definition, scope it to a handful of tools, run it under approval, and read the transcript afterward.

The details behind each part of the gateway, for when you are building rather than evaluating.

Declaring operations, input and output schemas, validation, namespacing, versioning, and the draft, approve, live lifecycle.

Machine identities for agents, per-seat identities for staff, how scopes resolve per request, and how revocation propagates.

Classifying operations T0 through T3, what evidence an approval carries, and how graduation and reversal are recorded.

Schedules, inbound email and documents, and system events. Run lifecycle, retries, exception handling, and manual invocation.

The run transcript format, per-call cost accounting, exporting records, and the queries auditors usually ask for.

Connecting Claude, Claude Cowork, or any MCP-compatible client to the gateway under the same permissions your agents use.

We would rather answer directly than have you guess. Write to us and we will point you at the right place, or write the page if it does not exist.