OpenRouter

Self-hosted LLM routing API. OpenAI-compatible /chat/completions endpoint that routes across OpenAI, Anthropic, and Google Gemini. Includes a user dashboard for API key management and credit tracking.

Stack

• Runtime: Bun
• API: Elysia (single process — auth, management, and LLM proxy on one port)
• Database: PostgreSQL + Prisma 7
• Frontend: React 19 + Tailwind CSS v4 + shadcn/ui (built to static files, served by Elysia)
• Monorepo: Turborepo

Quick start

bun install

# .env
DATABASE_URL=postgresql://user:password@localhost:5432/openrouter
JWT_SECRET=generate-a-random-secret
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GOOGLE_API_KEY=...

bun run --cwd packages/db prisma migrate deploy
bun run packages/db/seed.ts
bun dev

Open http://localhost:3001 — the frontend, backend API, and LLM proxy all run on the same server.

Deploy (Fly.io + Neon)

fly launch --no-deploy
fly postgres create --name openrouter-db
fly postgres attach openrouter-db
fly secrets set JWT_SECRET=<random> OPENAI_API_KEY=... ANTHROPIC_API_KEY=... GOOGLE_API_KEY=...
fly deploy --build-arg DATABASE_URL="<internal-db-url>"
fly ssh console -C "bun run --cwd packages/db prisma migrate deploy"
fly ssh console -C "bun run packages/db/seed.ts"

The free tier (3 shared VMs + 1GB Postgres) handles this comfortably.

API

POST /api/v1/chat/completions
Authorization: Bearer <api-key>

Non-streaming

{
  "model": "openai/gpt-4o",
  "messages": [{"role": "user", "content": "Hello"}]
}

Streaming

{
  "model": "openai/gpt-4o",
  "messages": [{"role": "user", "content": "Hello"}],
  "stream": true
}

Model format: company/model-name (e.g. openai/gpt-4o, anthropic/claude-3.5-sonnet, google/gemini-1.5-pro). Provider is selected randomly from mapped providers.

Credits

credits = (inputTokens * inputTokenCost + outputTokens * outputTokenCost) / 10

New users start with 1,000 credits. Credits are consumed per-request and tracked per API key.

License

MIT