AI Infrastructure · Developer PlatformR&D

UNIUN Inference — AI gateway

“Our own OpenRouter” — one API key and endpoint that routes to subscriptions, provider APIs, or local models.

UNIUN Inference — one API key routed to Claude, GPT, Gemini, Grok and local models

A unified inference gateway that abstracts every LLM provider behind one key — with metering, billing, failover and difficulty-based smart routing.

The problem

Every team wiring up multiple LLM providers repeats the same work: different APIs, auth, pricing, streaming quirks and lock-in. UNIUN Inference is a single abstraction between apps and all providers.

How we built it

Own / Run / Build — a forked Go llm-gateway (OpenAI + Anthropic frontends, routing, failover, OTel), a CLIProxyAPI sidecar that turns subscription logins into API endpoints, and a UNIUN layer for keys, metering, billing and policy. One repo, two containers, one docker-compose up.

What it does

OpenAI and Anthropic wire protocols, both first-class
A backend is a Provider config entry, not a code fork
Difficulty-cascade smart routing — cheap work to cheap sources
DB-backed API keys, per-token metering, rate limits and billing
OpenTelemetry metrics throughout

Next project

BillBridge OCR — bills to Excel / Tally

Point a camera at any bill and it lands as clean rows in Excel — or vouchers in Tally.

Open it