Gemini API Documentation (Protobuf Definitions)
This skill provides the official v1beta Protobuf definitions for the Google Generative Language API (Gemini). These files contain the absolute source of truth for API capabilities, request/response structures, and schema definitions.
When to use these references
When implementing or debugging Gemini API clients (such as in Rust, Go, or other languages), refer to these files for the exact structure of messages, enums, and RPC methods.
Core Services & Schemas
The .proto files are located in references/. Here is a guide on where to look for specific features:
Content Generation & Tools
generative_service.proto: Contains the core GenerateContentRequest and GenerateContentResponse. Look here for tool configurations, system instructions, and generation configurations.content.proto: Defines the Content, Part, FunctionCall, FunctionResponse, Tool, and ToolConfig structures. Essential for understanding how to construct multimodal payloads and function calling schemas.safety.proto: Defines HarmCategory, SafetySetting, and SafetyRating enums and messages.
Models & Metadata
model.proto: Defines the Model message, including properties like input_token_limit, output_token_limit, and supported generation methods.model_service.proto: RPCs for listing and getting models (GetModel, ListModels).
Context Caching
cached_content.proto: Defines the CachedContent resource for caching large prompts.cache_service.proto: RPCs for managing cached contents (CreateCachedContent, UpdateCachedContent, etc.).
File API (Multimodal Uploads)
file.proto: Defines the File resource representing uploaded media (video, images, documents).file_service.proto: RPCs for the File API (CreateFile, ListFiles, GetFile, DeleteFile).
Semantic Retrieval (RAG/AQA)
retriever.proto / retriever_service.proto: Definitions for Corpora, Documents, Chunks, and operations like Semantic Search and Attributed Question Answering (AQA).
Fine-tuning
tuned_model.proto: Definitions for tuning jobs and datasets.permission.proto / permission_service.proto: Access control for tuned models and corpora.
Legacy / PaLM Services
discuss_service.proto / text_service.proto / prediction_service.proto: Older APIs (like GenerateMessage or GenerateText), primarily for PaLM 2 models. Usually not needed for modern Gemini implementations.
How to use
If you need to know the exact field names, types, or enums (e.g., "Does ToolConfig take a string or an enum for function calling mode?"), use grep_search or view the relevant file in the references/ directory.
Example: If implementing function calling, check content.proto for FunctionCall and generative_service.proto for where tools is placed in the GenerateContentRequest.
1---2name: gemini-api3description: Provides the official Protobuf specifications and gRPC/REST service definitions for the Gemini API (Google Generative Language API v1beta). Use this skill when you need precise schema definitions for requests/responses (e.g., GenerateContent, Tools, FunctionCalling), multimodal content structures (Content, Part), context caching, model metadata, safety settings, or file uploads.4---56# Gemini API Documentation (Protobuf Definitions)78This skill provides the official v1beta Protobuf definitions for the Google Generative Language API (Gemini). These files contain the absolute source of truth for API capabilities, request/response structures, and schema definitions.910## When to use these references1112When implementing or debugging Gemini API clients (such as in Rust, Go, or other languages), refer to these files for the exact structure of messages, enums, and RPC methods.1314## Core Services & Schemas1516The `.proto` files are located in `references/`. Here is a guide on where to look for specific features:1718### Content Generation & Tools19- **`generative_service.proto`**: Contains the core `GenerateContentRequest` and `GenerateContentResponse`. Look here for tool configurations, system instructions, and generation configurations.20- **`content.proto`**: Defines the `Content`, `Part`, `FunctionCall`, `FunctionResponse`, `Tool`, and `ToolConfig` structures. Essential for understanding how to construct multimodal payloads and function calling schemas.21- **`safety.proto`**: Defines `HarmCategory`, `SafetySetting`, and `SafetyRating` enums and messages.2223### Models & Metadata24- **`model.proto`**: Defines the `Model` message, including properties like `input_token_limit`, `output_token_limit`, and supported generation methods.25- **`model_service.proto`**: RPCs for listing and getting models (`GetModel`, `ListModels`).2627### Context Caching28- **`cached_content.proto`**: Defines the `CachedContent` resource for caching large prompts.29- **`cache_service.proto`**: RPCs for managing cached contents (`CreateCachedContent`, `UpdateCachedContent`, etc.).3031### File API (Multimodal Uploads)32- **`file.proto`**: Defines the `File` resource representing uploaded media (video, images, documents).33- **`file_service.proto`**: RPCs for the File API (`CreateFile`, `ListFiles`, `GetFile`, `DeleteFile`).3435### Semantic Retrieval (RAG/AQA)36- **`retriever.proto` / `retriever_service.proto`**: Definitions for Corpora, Documents, Chunks, and operations like Semantic Search and Attributed Question Answering (AQA).3738### Fine-tuning39- **`tuned_model.proto`**: Definitions for tuning jobs and datasets.40- **`permission.proto` / `permission_service.proto`**: Access control for tuned models and corpora.4142### Legacy / PaLM Services43- **`discuss_service.proto` / `text_service.proto` / `prediction_service.proto`**: Older APIs (like `GenerateMessage` or `GenerateText`), primarily for PaLM 2 models. Usually not needed for modern Gemini implementations.4445## How to use4647If you need to know the exact field names, types, or enums (e.g., "Does `ToolConfig` take a string or an enum for function calling mode?"), use `grep_search` or view the relevant file in the `references/` directory.4849Example: If implementing function calling, check `content.proto` for `FunctionCall` and `generative_service.proto` for where `tools` is placed in the `GenerateContentRequest`.
stephen94125