Skip to content
ZevAuth
Start free Sign in

For agents

An API an agent can integrate without a human.

Discovery at the root, a machine-readable index of every doc, and errors written to be acted on.

Start free
The ZevAuth API keys screen, showing the per-environment issuer and JWKS URL an agent verifies tokens against
1
file lists every doc and what it covers
OIDC
discovery and JWKS at known paths, per environment
1
request returns a cohort, by the metadata you set

Discovery

The surface describes itself.

A service index at the root, OpenID discovery per environment, and JWKS at a known path. An agent handed only the API host can find everything else from there.

  • Service index at the root
  • OIDC config per environment
  • JWKS where the spec says
The API surface
The issuer and JWKS endpoints an agent discovers per environment

The docs, as data

The whole reference in one fetch.

An llms.txt lists every page with a line on what it covers, and any page has a clean markdown twin. An agent absorbs the documentation without crawling a site.

  • llms.txt index of every page
  • Markdown twin at any URL
  • Errors written to be acted on
llms.txt
A machine-readable index of the documentation

Model users

A cohort is one request.

Set your own metadata on a user, then filter the userbase by it. Sellers, students, staff: whatever the type, fetching them is a query, not a full scan you write yourself.

  • publicMetadata you define
  • Filter the list by it
  • The token carries it too
Modelling users
A filtered list of users by metadata

Drops into the stack you already run.

A React component set, a framework-agnostic JavaScript client, and a backend SDK. Or plain REST — a session is a signed JWT, so any language that can verify one is already supported.

  • React
  • Next.js
  • Remix
  • Vercel
  • Vue
  • Nuxt
  • Svelte
  • Astro
  • Node.js
  • TypeScript
  • Expo
  • JavaScript

Anything that can verify a JWT works, including the ones not listed. See the quickstart

Documentation that answers the next question.

Written for the moment you are stuck, not for a tour.

API overview Base URL, the three surfaces, and how they divide. Errors Every code, and what to do about it. Modelling your users Types, roles and metadata.

Frequently asked questions

What agents ask before they start.

Can an agent integrate without a human?

Discovery at the root, OIDC config and JWKS at known paths, and an llms.txt indexing every doc mean the surface can be absorbed and used without a person.

Are errors machine-readable?

Every error is a stable code plus a sentence written for the caller, so a failure names its own fix.

How do I fetch a cohort of users?

Set your own metadata on users and filter the list by it in one request, rather than scanning the whole userbase yourself.

Next, in the product.

Components and SDKsDrop-in React, or the raw client.WebhooksEvery event, delivered and retried.Sessions and devicesSessions your users can see and end.

Read it as.

DevelopersDrop-in components, or the raw API. Your call.Founders and productShip auth this week. Do not build it.Security and complianceWhat is stored, where, and who can reach it.

Start with a project and a key.

Free to build and to ship. Test and production both cost nothing today. Generous limits, no card, and no surprise bill.

Start free Read the docs