QuantumNous/new-api

New API – A Self‑Hosted AI Gateway

What it is – New API is an open‑source, self‑hosted gateway that sits between your applications (or AI agents) and various large‑language‑model providers. It presents a single, consistent OpenAI‑compatible HTTP API while handling routing, usage tracking, access control, pricing, and multi‑model management.

Key capabilities

• Multi‑provider support – Connect upstream services such as OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock, Vertex AI, DeepSeek, Qwen, etc.
• Unified endpoints – Exposes familiar OpenAI‑style routes (/v1/chat/completions, /v1/messages, /v1/embeddings, image/audio generation, WebSocket realtime, etc.) plus a plugin task system.
• Routing & fail‑over – Define channel priorities, weights, retries, and affinity to automatically balance or switch providers.
• Usage & cost accounting – Track token usage, cache accounting, quotas, subscriptions, and expression‑based pricing per tier.
• Fine‑grained access control – Users, groups, API‑key scopes, OAuth/OIDC, passkeys, 2FA, and session management.
• Extensible task plugins – JavaScript plugins let you add custom asynchronous jobs (e.g., image processing, video generation) with status endpoints.
• Web console – A multilingual React‑based UI for configuring models, viewing dashboards, managing users, and testing the API.

How to get started

1. Run with Docker – A single‑command Docker run spins up New API with an SQLite store.

   mkdir -p data
   docker run --name new-api -d --restart unless-stopped \
     -p 127.0.0.1:3000:3000 \
     -e TZ=Asia/Shanghai \
     -v "$(pwd)/data:/data" calciumion/new-api:latest
   

2. Open http://localhost:3000, run the setup wizard, add an upstream channel (e.g., your OpenAI key), and create an API key.
3. Point your client to http://localhost:3000/v1 and use the issued key just like you would with the original provider.

Deployment options

• Docker‑Compose – Includes PostgreSQL + Redis (or MySQL/ClickHouse) for production‑grade persistence and rate‑limiting.
• Configurable storage – SQLite, MySQL, PostgreSQL, optional separate log DB, optional Redis cache.
• Environment variables control DB connections, secrets, cookie security, trusted proxies, etc.
• Multi‑node setups share the main DB and secret keys; Redis can be used for shared rate limits.

Development

• Backend: Go + Gin framework.
• Frontend: React 19, TypeScript, Tailwind 4, built with Bun.
• relaykit – a Go module that normalises the four major text protocols.
• Plugins live under plugins/tasks/; write JavaScript plugins following the Task Plugin API v1.
• Build steps: bun install && bun run build for the UI, then go run . for the server, or use the provided make dev workflow.

Documentation & community

• Full docs: https://docs.newapi.ai/en/docs (installation, API reference, guides).
• Community FAQ, discussion, and issue tracker on GitHub.
• Contributions welcome for translations, new provider adapters, tests, and plugins.

License – AGPL‑v3 with additional attribution requirements.

---

In short, New API lets teams host their own “one‑stop shop” for LLM access, giving them control over costs, security, and provider choice while keeping the familiar OpenAI‑style API.