Bring Your Own Key (BYOK)

Replace YOUR_NANOGPT_API_KEY and choose one of your saved model provider keys: openai | openai-responses | anthropic | chutes | aws | groq | sambanova | vercel | novita | meganova | xiaomi | siliconflow | wafer | akash | google | google-vertex | zai | gerra | gmicloud | cerebras | deepinfra. Web search keys are used only for web search endpoints and enhancements.

curl -X POST https://nano-gpt.com/api/v1/chat/completions \
  -H "x-api-key: YOUR_NANOGPT_API_KEY" \
  -H "x-use-byok: true" \
  # Optional but recommended: force provider mapping when in doubt \
  -H "x-byok-provider: openai" \
  -H "content-type: application/json" \
  -d '{
    "model": "gpt-4o-mini",
    "messages": [
      {"role": "user", "content": "Say hello from BYOK"}
    ]
  }'

We charge a 5% markup on the normal at-cost API rates. Your provider bills you directly; we only charge the markup.

If the model that you use is free via the original provider where you use it from we base our 5% markup on what the lowest rate would be if you used the paid version of it. If you for example use an API key that has free requests for gemini-2.5-pro, we will charge the 5% markup based on regular gemini-2.5-pro pricing.

BYOK is supported on the OpenAI-compatible endpoint /api/v1/chat/completions. Web chat can use saved provider keys for normal text chat when saved-key mode is enabled; the standard provider selector remains platform-billed.

Web search keys are used for web search only (e.g., /api/web and web-search enhancements). They are not chat model providers.

No. It’s optional. We auto-map based on the model. Setting it is recommended if you want to force a specific provider.

Add your Google AI Studio provider key above, then set the header x-byok-provider: google when calling /api/v1/chat/completions. This forces BYOK routing through Google AI Studio for Gemini models.

Add your Google Vertex service account JSON above with a location, then set x-byok-provider: google-vertex for Gemini BYOK chat requests.

We prefer provider-reported usage tokens when available (streaming or final usage). If unavailable, we estimate conservatively.

Saved provider keys are encrypted at rest. We only store the last 4 characters for display and never send keys back to the client.