w/01 · ai infrastructure · hypery.ai

Hypery

the identity and payments layer for ai products.

what

identity, ai gateway & billing for ai apps

codebase

158k lines · 665 commits

platform

web · react sdk · rest

stack

next.js 16 · mongodb · stripe · webgpu

move to bend the light

overview

Every AI product ends up rebuilding the same foundation before it can ship the interesting part: sign-in, access to models, metering, spending limits, a way to get paid. Hypery is that foundation as one product.

Developers drop in a React SDK for auth and checkout, point any OpenAI client at the Hypery gateway, and charge per use — with limits enforced before a single provider is called.

200+

models behind one endpoint

sign-in methods + oauth provider

100%

openai sdk compatible

the homepage backdrop is a live WebGPU shader that follows the cursor

the challenge

AI apps fail in boring ways: a stream dies halfway and nobody knows whether to charge; a webhook goes missing and credits get granted twice; a retry double-bills a customer.

Hypery had to make identity and money feel instant and trustworthy — and make the brand feel like the frontier it serves, not another dashboard.

01 · authentication

sign-in for ai apps, done properly.

Email and password, magic links, passkeys, TOTP two-factor, enterprise SSO and social sign-in with Google, GitHub and Apple — built on better-auth.

Hypery is also a full OAuth 2.0 provider: authorization code with PKCE (S256 only), scopes, a consent screen and co-branded sign-in, so apps can offer “Sign in with Hypery”.

→drop-in <SignIn/>, <UserButton/>, <AuthModal/>
→scoped api keys & team invitations
→popup flows that close themselves when another tab finishes

02 · ai gateway

one endpoint, every model.

OpenAI-compatible /api/v1 routes for chat, embeddings, images, speech, transcription, realtime sessions and predictions — routed across OpenRouter, Replicate, Hugging Face and the Vercel AI Gateway.

Streams reconnect when a first byte stalls (with grace for reasoning models), and an abort sends a clean SSE error frame instead of a hung socket.

→function calling, vision & prompt caching
→model catalog synced hourly
→works with LangChain, Continue.dev and the Vercel AI SDK

03 · billing

money that can't double-count.

Pay-as-you-go by default — usage is metered straight to Stripe — or a credit ledger with atomic reserve and finalize, auto top-up, grants and per-app spending caps checked before the provider is ever called.

Developers can sell their own subscription plans to their users, with Stripe Connect handling payouts and a platform fee.

→idempotency keys claimed atomically per team
→credits granted once per payment intent
→reconciliation crons for missed webhooks & stale reservations

04 · models

switch models with one string.

Chat and image models from OpenRouter, Replicate and Hugging Face behind the same endpoint: change the model name, keep your code. Every page of the site gets its own variation of the shader — streaks, aurora, beams, waves, warp and more.

05 · pricing

$0 a month. pay for what you use.

No commitment and no seat tax: pricing is usage, with spending limits per app and per team so a runaway loop can never become a runaway bill.

architecture

a modular monolith with very careful money.

One Next.js 16 app serves the marketing site, docs, dashboard, OAuth server and the gateway API — deployed on Vercel with crons doing the reconciliation work a queue would.

01
sdk
@hyperyai/sdk — React components + hooks, PKCE popup, SSE parsing
→
02
oauth & auth
better-auth, passkeys, SSO, consent, co-branded sign-in
→
03
gateway
OpenAI-compatible API → provider routing → streaming with reconnects
→
04
metering
reserve → call → finalize, rollback on failure, idempotent retries
→
05
stripe
meters, top-ups, subscriptions, Connect payouts, webhook backstops
→
06
data
MongoDB (native driver), Redis rate limits, S3 media, 11 Vercel crons

the developer experience

point any openai client at it.

adapted from the Hypery quickstart — the gateway speaks the OpenAI API, so existing code just works.

quickstart.ts

1import OpenAI from 'openai'
2 
3const client = new OpenAI({
4  apiKey: process.env.GATEWAY_API_KEY,
5  baseURL: 'https://hypery.ai/api/v1',
6})
7 
8const res = await client.chat.completions.create({
9  model: 'anthropic/claude-sonnet-4.5',
10  messages: [{ role: 'user', content: 'hello, hypery' }],
11  stream: true,
12})

brand & design system

A dark-first, navy-tinted system built in oklch, with a brand spectrum that runs from cyan through violet to warm orange — the colors of the icon, animated by the WebGPU backdrop.

accent

#00b6d1

blue

#3672e9

violet

#9c5eef

warm

#fda856

deep

#0c2352

night

#07101c