One API — Unified LLM Gateway for OpenAI, Claude, and Gemini

Introduction

One API is a self-hosted LLM API gateway that lets you access OpenAI, Anthropic Claude, Google Gemini, and dozens of other providers through a single OpenAI-compatible endpoint. Teams use it to centralize API key management, enforce usage quotas, and switch between providers without changing application code.

What One API Does

• Aggregates 30+ LLM providers behind one unified API endpoint compatible with the OpenAI format
• Manages API keys, user tokens, and per-user quota limits through a web dashboard
• Load-balances requests across multiple provider channels with automatic failover
• Tracks token usage and costs per user, per channel, and per model
• Supports streaming responses, function calling, and vision models across providers

Architecture Overview

One API is a Go application backed by SQLite or MySQL. The web dashboard is built with React. Incoming requests are routed through a channel selection layer that picks the best available provider based on priority, weight, and health status. Each channel wraps a specific provider adapter that translates between the OpenAI format and the native API format.

Self-Hosting & Configuration

• Deploy via Docker, Docker Compose, or a single binary on Linux, macOS, or Windows
• Set the database with SQL_DSN for MySQL or use the default embedded SQLite
• Add provider channels through the admin UI with API keys and model mappings
• Create user tokens with optional rate limits, expiration, and quota caps
• Enable Redis for distributed rate limiting in multi-instance deployments

Key Features

• OpenAI-compatible endpoint means zero code changes for existing applications
• Built-in billing system tracks token consumption and supports recharge codes
• Channel priority and weight system for cost optimization across providers
• Automatic retry and failover when a provider returns errors or timeouts
• Model mapping lets you expose custom model names that route to any backend

Comparison with Similar Tools

• LiteLLM — Python-based proxy; One API offers a full admin UI with quota management
• OpenRouter — Managed service; One API is fully self-hosted with no third-party dependency
• Portkey AI Gateway — Focused on observability; One API emphasizes multi-tenant billing
• API Gateway (Kong/APISIX) — General-purpose; One API is purpose-built for LLM token economics
• Helicone — Logging-focused proxy; One API adds load balancing and provider aggregation

FAQ

Q: Which LLM providers does One API support? A: OpenAI, Azure OpenAI, Anthropic Claude, Google Gemini, AWS Bedrock, Mistral, DeepSeek, Ollama, and many more. New providers are added regularly.

Q: Can I use One API with existing OpenAI SDK code? A: Yes. Point the base_url to your One API instance and use the issued token as the API key. No other changes are needed.

Q: Does it support streaming? A: Yes. Streaming, function calling, and multimodal (vision) requests are fully supported across compatible providers.

Q: How does quota management work? A: Admins set token quotas per user. Each request deducts from the quota based on actual token usage reported by the provider.

Sources

• https://github.com/songquanpeng/one-api
• https://openai.justsong.cn/