gproxy-protocol

中文文档 / Chinese

Wire-format types and cross-protocol transforms for the three major LLM APIs — Anthropic Claude, OpenAI (ChatCompletions + Responses), and Google Gemini. Zero HTTP dependencies, pure serde data + typed converters.

gproxy-protocol is the protocol types crate of the gproxy project. It covers request, response, streaming event, and shared types for the Claude, OpenAI, and Gemini APIs, and provides cross-protocol conversions through the transform module.

Quick start

Typed cross-protocol conversion via TryFrom:

use gproxy_protocol::claude::create_message::request::ClaudeCreateMessageRequest;
use gproxy_protocol::openai::create_chat_completions::request::OpenAiChatCompletionsRequest;

fn claude_to_openai(
    req: ClaudeCreateMessageRequest,
) -> Result<OpenAiChatCompletionsRequest, gproxy_protocol::transform::TransformError> {
    OpenAiChatCompletionsRequest::try_from(req)
}

Or runtime-keyed dispatch when source / destination only come from config:

use gproxy_protocol::kinds::{OperationFamily, ProtocolKind};
use gproxy_protocol::transform::dispatch::transform_request;

let out: Vec<u8> = transform_request(
    OperationFamily::GenerateContent, ProtocolKind::Claude,
    OperationFamily::GenerateContent, ProtocolKind::OpenAiChatCompletion,
    body_bytes,
)?;

SSE → NDJSON framing rewriter for gateway passthrough:

use gproxy_protocol::stream::SseToNdjsonRewriter;

let mut rewriter = SseToNdjsonRewriter::default();
let out = rewriter.push_chunk(b"data: {\"text\":\"hi\"}\n\n");
assert_eq!(out, b"{\"text\":\"hi\"}\n");

Public Entry Points

Entry Point	Source	Description
`pub mod claude`	`src/lib.rs`	Claude protocol module.
`pub mod openai`	`src/lib.rs`	OpenAI protocol module.
`pub mod gemini`	`src/lib.rs`	Gemini protocol module.
`pub mod stream`	`src/lib.rs`	SSE / NDJSON stream processing utilities.
`pub mod transform`	`src/lib.rs`	Cross-protocol transformation matrix.
`pub use kinds::{OperationFamily, ProtocolKind}`	`src/lib.rs`	Shared operation and protocol enums used across modules.

Supported Protocols

Protocol	Module	Description
Claude	`gproxy_protocol::claude`	Types related to the Anthropic Messages API, files, models, and token counting.
OpenAI	`gproxy_protocol::openai`	Types for Chat Completions, Responses, Embeddings, Images, Models, and related APIs.
Gemini	`gproxy_protocol::gemini`	Types for GenerateContent, StreamGenerateContent, Embeddings, Batch Embeddings, Models, and the Live API.

Protocol Operation List

Claude

Operation Module	Primary Endpoint / Purpose	Typical Public Types
`create_message`	`POST /v1/messages`	`ClaudeCreateMessageRequest`, `ClaudeCreateMessageResponse`, `ClaudeStreamEvent`, `BetaMessage`
`count_tokens`	`POST /v1/messages/count_tokens`	`ClaudeCountTokensRequest`, `ClaudeCountTokensResponse`, `BetaMessageTokensCount`
`model_list`	`GET /v1/models`	`ClaudeModelListRequest`, `ClaudeModelListResponse`
`model_get`	`GET /v1/models/{model_id}`	`ClaudeModelGetRequest`, `ClaudeModelGetResponse`, `BetaModelInfo`
`file_upload`	`POST /v1/files`	`ClaudeFileUploadRequest`, `ClaudeFileUploadResponse`, `FileMetadata`
`file_list`	`GET /v1/files`	`ClaudeFileListRequest`, `ClaudeFileListResponse`
`file_download`	`GET /v1/files/{file_id}/content`	`ClaudeFileDownloadRequest`, `ClaudeFileDownloadResponse`
`file_get`	`GET /v1/files/{file_id}`	`ClaudeFileGetRequest`, `ClaudeFileGetResponse`, `FileMetadata`
`file_delete`	`DELETE /v1/files/{file_id}`	`ClaudeFileDeleteRequest`, `ClaudeFileDeleteResponse`, `DeletedFile`

OpenAI

Operation Module	Primary Endpoint / Purpose	Typical Public Types
`create_chat_completions`	`POST /v1/chat/completions`	`OpenAiChatCompletionsRequest`, `OpenAiChatCompletionsResponse`, `ChatCompletion`, `ChatCompletionChunk`
`create_response`	`POST /v1/responses`	`OpenAiCreateResponseRequest`, `OpenAiCreateResponseResponse`, `ResponseBody`, `ResponseStreamEvent`
`create_response::websocket`	Responses WebSocket connection and messages	`OpenAiCreateResponseWebSocketConnectRequest`, `OpenAiCreateResponseWebSocketClientMessage`, `OpenAiCreateResponseWebSocketServerMessage`
`compact_response`	`POST /v1/responses/{id}/compact`	`OpenAiCompactRequest`, `OpenAiCompactResponse`, `CompactedResponseOutputItem`
`count_tokens`	`POST /v1/responses/input_tokens/count`	`OpenAiCountTokensRequest`, `OpenAiCountTokensResponse`
`embeddings`	`POST /v1/embeddings`	`OpenAiEmbeddingsRequest`, `OpenAiEmbeddingsResponse`, `OpenAiCreateEmbeddingResponse`
`create_image`	`POST /v1/images/generations`	`OpenAiCreateImageRequest`, `OpenAiCreateImageResponse`, `ImageGenerationStreamEvent`
`create_image_edit`	`POST /v1/images/edits`	`OpenAiCreateImageEditRequest`, `OpenAiCreateImageEditResponse`, `ImageEditStreamEvent`
`model_list`	`GET /v1/models`	`OpenAiModelListRequest`, `OpenAiModelListResponse`, `OpenAiModelList`
`model_get`	`GET /v1/models/{model}`	`OpenAiModelGetRequest`, `OpenAiModelGetResponse`, `OpenAiModel`

Gemini

Operation Module	Primary Endpoint / Purpose	Typical Public Types
`generate_content`	`POST models/{model}:generateContent`	`GeminiGenerateContentRequest`, `GeminiGenerateContentResponse`, `gemini::generate_content::response::ResponseBody`
`stream_generate_content`	`POST models/{model}:streamGenerateContent`	`GeminiStreamGenerateContentRequest`, `GeminiStreamGenerateContentResponse`, `GeminiNdjsonChunk`, `GeminiSseChunk`
`count_tokens`	`POST models/{model}:countTokens`	`GeminiCountTokensRequest`, `GeminiCountTokensResponse`
`embeddings`	`POST models/{model}:embedContent`	`GeminiEmbedContentRequest`, `GeminiEmbedContentResponse`, `GeminiContentEmbedding`
`batch_embed_contents`	`POST models/{model}:batchEmbedContents`	`GeminiBatchEmbedContentsRequest`, `GeminiBatchEmbedContentsResponse`, `BatchRequestItem`
`model_list`	`GET models`	`GeminiModelListRequest`, `GeminiModelListResponse`
`model_get`	`GET models/{model}`	`GeminiModelGetRequest`, `GeminiModelGetResponse`, `GeminiModelInfo`
`live`	Live API / BidiGenerateContent WebSocket	`GeminiLiveConnectRequest`, `GeminiBidiGenerateContentClientMessage`, `GeminiBidiGenerateContentServerMessage`

Cross-Protocol Transformation Matrix

Notes:

In the source tree, OpenAI is split into two target or source shapes: openai_chat_completions and openai_response.
The table below is organized by whether the corresponding submodule exists under src/transform/; it does not include conversions within the same protocol.

Operation	Claude ↔ OpenAI	Claude ↔ Gemini	OpenAI ↔ Gemini	Evidence
`model_list`	Bidirectional	Bidirectional	Bidirectional	`transform/{claude,openai,gemini}/model_list/*`
`model_get`	Bidirectional	Bidirectional	Bidirectional	`transform/{claude,openai,gemini}/model_get/*`
`count_tokens`	Bidirectional	Bidirectional	Bidirectional	`transform/{claude,openai,gemini}/count_tokens/*`
`generate_content`	Bidirectional	Bidirectional	Bidirectional	`transform/claude/generate_content/`, `transform/openai/generate_content/`, `transform/gemini/generate_content/*`
`stream_generate_content`	Bidirectional	Bidirectional	Bidirectional	`transform/claude/stream_generate_content/`, `transform/openai/stream_generate_content/`, `transform/gemini/stream_generate_content/*`
`embeddings`	Not supported	Not supported	Bidirectional	`transform/openai/embeddings/gemini`, `transform/gemini/embeddings/openai`
`compact`	One-way `OpenAI → Claude`	Not supported	One-way `OpenAI → Gemini`	`transform/openai/compact/{claude,gemini}`
`create_image`	Not supported	Not supported	One-way `OpenAI → Gemini`	`transform/openai/create_image/gemini`
`create_image_edit`	Not supported	Not supported	One-way `OpenAI → Gemini`	`transform/openai/create_image_edit/gemini`
`websocket` / `live`	Not supported	Not supported	Each side only provides its own HTTP ↔ WebSocket bridge; no Claude interop is provided	`transform/openai/websocket/`, `transform/gemini/websocket/`
`file_*`	Not supported	Not supported	Not supported	There is no file-operation directory under `transform/`

Key Public Types

Category	Type	Location	Purpose
Root type	`OperationFamily`	`src/kinds.rs`	Protocol-agnostic operation family enum.
Root type	`ProtocolKind`	`src/kinds.rs`	Shared protocol enum used by routing, transforms, and provider dispatch.
Stream utility	`SseToNdjsonRewriter`	`src/stream.rs`	Incremental SSE → NDJSON rewriter.
Transform	`TransformError`	`src/transform/utils.rs`	Cross-protocol transform error.
Transform	`TransformResult<T>`	`src/transform/utils.rs`	Type alias for cross-protocol transform results.
Claude	`ClaudeCreateMessageRequest`	`src/claude/create_message/request.rs`	Claude message creation request.
Claude	`ClaudeCreateMessageResponse`	`src/claude/create_message/response.rs`	Claude message creation response enum.
Claude	`ClaudeStreamEvent`	`src/claude/create_message/stream.rs`	Claude streaming event.
Claude	`BetaMessage`	`src/claude/create_message/types.rs`	Claude message response body.
Claude	`ClaudeCountTokensRequest`	`src/claude/count_tokens/request.rs`	Claude token counting request.
Claude	`BetaMessageTokensCount`	`src/claude/count_tokens/types.rs`	Claude token counting result.
Claude	`BetaModelInfo`	`src/claude/types.rs`	Claude model metadata.
Claude	`FileMetadata`	`src/claude/types.rs`	Claude file metadata.
OpenAI	`OpenAiChatCompletionsRequest`	`src/openai/create_chat_completions/request.rs`	Chat Completions request.
OpenAI	`ChatCompletion`	`src/openai/create_chat_completions/types.rs`	Full Chat Completions response body.
OpenAI	`ChatCompletionChunk`	`src/openai/create_chat_completions/stream.rs`	Chat Completions streaming chunk.
OpenAI	`OpenAiCreateResponseRequest`	`src/openai/create_response/request.rs`	Responses API request.
OpenAI	`OpenAiCreateResponseResponse`	`src/openai/create_response/response.rs`	Responses API response enum.
OpenAI	`ResponseStreamEvent`	`src/openai/create_response/stream.rs`	Responses API streaming event.
OpenAI	`OpenAiCreateResponseWebSocketConnectRequest`	`src/openai/create_response/websocket/request.rs`	Responses WebSocket connection request.
OpenAI	`OpenAiEmbeddingsRequest`	`src/openai/embeddings/request.rs`	Embeddings request.
OpenAI	`OpenAiCreateEmbeddingResponse`	`src/openai/embeddings/types.rs`	Embeddings response body.
OpenAI	`OpenAiCreateImageRequest`	`src/openai/create_image/request.rs`	Image generation request.
OpenAI	`OpenAiCreateImageEditRequest`	`src/openai/create_image_edit/request.rs`	Image edit request.
OpenAI	`OpenAiCompactRequest`	`src/openai/compact_response/request.rs`	Response compaction request.
OpenAI	`OpenAiModel`	`src/openai/types.rs`	OpenAI model object.
OpenAI	`OpenAiModelList`	`src/openai/types.rs`	OpenAI model list body.
Gemini	`GeminiGenerateContentRequest`	`src/gemini/generate_content/request.rs`	GenerateContent request.
Gemini	`GeminiGenerateContentResponse`	`src/gemini/generate_content/response.rs`	GenerateContent response enum.
Gemini	`GeminiStreamGenerateContentRequest`	`src/gemini/stream_generate_content/request.rs`	StreamGenerateContent request.
Gemini	`GeminiNdjsonChunk`	`src/gemini/stream_generate_content/stream.rs`	NDJSON streaming chunk.
Gemini	`GeminiSseChunk`	`src/gemini/stream_generate_content/stream.rs`	SSE streaming chunk.
Gemini	`GeminiCountTokensRequest`	`src/gemini/count_tokens/request.rs`	CountTokens request.
Gemini	`GeminiEmbedContentRequest`	`src/gemini/embeddings/request.rs`	Single embedding request.
Gemini	`GeminiBatchEmbedContentsRequest`	`src/gemini/batch_embed_contents/request.rs`	Batch embedding request.
Gemini	`GeminiLiveConnectRequest`	`src/gemini/live/request.rs`	Live API connection request.
Gemini	`GeminiBidiGenerateContentClientMessage`	`src/gemini/live/types.rs`	Live API client message.
Gemini	`GeminiBidiGenerateContentServerMessage`	`src/gemini/live/types.rs`	Live API server message.
Gemini	`GeminiModelInfo`	`src/gemini/types.rs`	Gemini model metadata.

License

Licensed under the MIT License.

gproxy-protocol 1.0.20