Recommended reading order
For a new integration:- Create or select an API key in the console.
- Call the model list and confirm the model IDs available to that key.
- Choose an interface for your workload: text, image, video, audio, or files.
- Run a minimal request before adding streaming, tools, reference images, or task polling.
- Review usage and charges in the console.
Basic information
Production base URL:x-api-key. Gemini native requests may use x-goog-api-key or ?key=. Use Bearer authentication unless a client requires a native header.
Handling results
Response shapes depend on the interface:- Text usually appears in
choices[0].message.contentor the Responses output array. - Image results may be returned as a URL or base64 data.
- Video and music requests usually create a task first, then return a task ID for polling.
- File APIs have separate upload, retrieve, download, and delete operations.
AI-readable documentation
The documentation can be read directly by AI tools:- Copy as Markdown converts a page into Markdown for LLM context.
skill.mdgives agents the integration order and API boundaries.- MCP documentation describes the current self-hosted deployment boundary.