73 paid APIs.
$0.00 spent.

You're building an AI feature. Every test run costs money — $0.003 per GPT-4 completion, but you're testing 200 times a day while iterating on the prompt. By Friday, your dev costs are higher than the hosting bill. And that's just OpenAI. Factor in Stripe test charges, Twilio SMS tests, SendGrid email sends — the development cost of building against paid APIs is real.

Donk runs a local mock server that replaces 73 API providers with zero-cost, instant, realistic responses. Your server calls api.openai.com? Donk intercepts it via HTTP_PROXY and returns a mock completion with proper schema, streaming tokens over SSE just like the real API. Your browser calls api.stripe.com via fetch()? The gateway intercepts it, injects CORS headers, and returns a mock payment intent.

All 73 Providers

How Interception Works

• Server-side — HTTP_PROXY injected into your backend process. Every outgoing request to mocked hostnames redirected to the mock server. Even hardcoded SDKs.
• Browser-side — The gateway intercepts fetch() calls to mocked API hostnames. CORS headers injected. OPTIONS preflight handled.
• Per-provider control — Mock OpenAI but let Stripe hit the real API. Toggle per provider, per project.
• SSE streaming — OpenAI and Anthropic completions stream token by token, just like the real API
• Custom routes — Add your own mock endpoints for edge-case testing
• Request log with timestamps and matched routes