Unified Gateway
One OpenAI / Anthropic / Gemini compatible endpoint, so customers stop juggling provider keys and SDK adapters.
RELAY · RLAB.WORKS
Relay
B2B AI API relay & billing platform for teams.
Relay is a self-operated B2B AI API relay. It fronts OpenAI, Anthropic, and Gemini behind one compatible gateway — companies sign up, invite teammates, share a prepaid balance, and get per-member keys with per-token metering and transparent invoices.
30B+
Tokens routed (first month)
400+
Seed users
9
Upstream endpoints
CORE CAPABILITIES
A turnkey AI gateway for your customers
From signup to teammate invites to first API call, Relay productizes the entire B2B AI onboarding path — operators only configure models and pricing.
Organizations & Teams
Owner / Admin / Member roles with a shared prepaid balance across the org and an independent gateway key per member.
Prepaid Balance & Billing
Real-time per-token deduction with traceable invoices for every call; platform admins top up customer balances manually.
Model Routing
Battle-tested routing and credential-pool modules selected from SiftGate OSS, with multi-provider load balancing and failover.
Platform Admin
Admins configure available models, pricing, and customer credits; debug data is encrypted at rest and every op is auditable.
Production-Ready
One Docker Compose stack with Caddy-managed TLS — the entire platform runs on a single VPS.
INFRASTRUCTURE
An engineering foundation built for production traffic
Every layer — from service isolation to data consistency to deployment — is designed for high-throughput token metering and org-level collaboration. No over-engineering, no shortcuts.
Services
NestJS 11 双服务
admin-api and gateway run as isolated services — independently scaled, independently deployed, fault-contained.
Data & Metering
PostgreSQL 16 + Redis 7
PostgreSQL handles strongly-consistent writes for orgs, balances, and invoices; Redis meters tokens with millisecond precision.
Console
React 19 + Vite 6 + Tailwind 4
Customer console and platform admin share one modern frontend — instant builds, immediate feedback, zero operational friction.
Infrastructure
Docker Compose + Caddy
The entire production system is orchestrated in a single compose file, with Caddy managing TLS end-to-end — infrastructure complexity kept to a minimum.