one-api

Why we included this project

Most teams that build on LLMs eventually hit the same wall: every provider ships its own SDK and request format, and wiring them together eats time that should go into the product. One API removes that layer by standing in front of OpenAI, Azure, Anthropic, Google, DeepSeek, and a range of Chinese providers, and exposing them all through one OpenAI-compatible endpoint, so swapping or adding a backend doesn't touch your application code. It also covers the operational chores that usually get hacked together in-house: load balancing across channels, token and quota tracking, per-user and per-group rate limits, usage accounting, and redemption codes if you want to resell access. A single Go binary or Docker image gets it running, and multi-machine deployments scale it up for heavier traffic. If you manage keys for a team or hand out API access to internal users or customers, the admin UI and management API spare you a pile of custom plumbing.