Gemini API

Native Gemini paths on Omniall AI:

Action	Endpoint
Non-stream	`POST /v1beta/models/{model}:generateContent`
Stream	`POST /v1beta/models/{model}:streamGenerateContent`

Auth: Authorization: Bearer sk-..., or x-goog-api-key / ?key=. Many Gemini models are also callable via OpenAI-compatible POST /v1/chat/completions when exposed that way in Model Square.

Scenario	How
Text chat	`contents[].parts[].text`
Image understanding	`inlineData` (base64) parts
File analysis	`fileData.fileUri` or Chat `type: file`
Web / grounding	`tools: [{ "googleSearch": {} }]` when supported
Agent	`tools[].functionDeclarations`
Image generation	`generationConfig.responseModalities` includes `IMAGE`
Banana 2.1 reference gen	`gemini-nano-banana-2.1`; `inlineData` supports image / `application/pdf` / `video/mp4`; Chat uses `image_url` / `file` / `video_url`
Banana 2.1 Thinking	Native `generationConfig.thinkingConfig.thinkingLevel`; Chat `extra_body.google.thinking_config.thinking_level`; levels `minimal` / `medium` (default) / `high`

Image details: Gemini native format, Chat image generation. Official docs: generateContent