feat(proxy): Gemini Native API proxy integration by yovinchen · Pull Request #1918 · farion1231/cc-switch

yovinchen · 2026-04-06T04:42:10Z

Summary

Adds full Gemini Native (generateContent) API support to the local proxy server, enabling Claude Code to transparently use Gemini models as upstream providers with bidirectional protocol translation.

Closes #1899

Motivation

Currently the proxy only supports OpenAI-compatible and Anthropic-native upstream formats. Users who want to use Google's Gemini models must go through an OpenAI-compatible shim, which loses Gemini-specific features (native tool calling, thought signatures, safety feedback). This PR adds first-class Gemini Native support so the proxy can speak Gemini's own generateContent / streamGenerateContent protocol directly.

Changes

New Modules (Backend)

File	Purpose
`proxy/gemini_url.rs`	URL normalization — converts legacy OpenAI-compatible paths to canonical Gemini `models/*:generateContent` format. Supports both structured (official) and opaque (custom relay) full URLs.
`proxy/providers/gemini_schema.rs`	Tool schema conversion — maps Anthropic JSON Schema tool definitions to Gemini `FunctionDeclaration`, intelligently choosing between `parameters` (restricted) and `parametersJsonSchema` (full JSON Schema) based on complexity.
`proxy/providers/gemini_shadow.rs`	Session state shadow store — thread-safe LRU cache (200 sessions × 64 turns) tracking assistant content snapshots and tool call metadata for thought signature replay and tool result name resolution.
`proxy/providers/streaming_gemini.rs`	Streaming converter — transforms Gemini `streamGenerateContent?alt=sse` chunks into Anthropic-style SSE events (`message_start`, `content_block_delta`, `message_delta`, `message_stop`).
`proxy/providers/transform_gemini.rs`	Bidirectional request/response transformation — full Anthropic Messages API ↔ Gemini Native `generateContent` conversion including system prompts, messages, tools, tool results, tool choice, and usage metadata.

Modified Modules (Backend)

forwarder.rs — Endpoint rewriting (/v1/messages → Gemini generateContent), conditional Anthropic header filtering, query parameter merging, shadow store propagation.
handlers.rs — Gemini streaming/non-streaming response routing, tool schema hint extraction from original request body.
providers/claude.rs — Gemini Native format detection, provider type mapping (Gemini / GeminiCli), Google OAuth + API key auth strategy, request transformation delegation.
server.rs — GeminiShadowStore initialization in ProxyState.
sse.rs — Extracted shared take_sse_block() helper with CRLF delimiter support.
stream_check.rs — Gemini Native health check: URL resolution, auth header generation, request body transformation.

Frontend

Provider Presets — New "Gemini Native" preset (generativelanguage.googleapis.com, default models: gemini-2.5-pro / gemini-2.5-flash).
Form Fields — gemini_native option in API format selector with conditional hints for base URL and full URL mode.
Types — "gemini_native" added to ClaudeApiFormat union.
i18n — Chinese, English, Japanese translations for all new UI strings.

Architecture

Transformation Pipeline

┌─────────────────────┐     ┌──────────────────────┐     ┌─────────────────────┐
│  Claude Code CLI    │────▶│  CC Switch Proxy      │────▶│  Gemini Native API  │
│  (Messages API)     │◀────│  (Bidirectional Xform)│◀────│  (generateContent)  │
└─────────────────────┘     └──────────────────────┘     └─────────────────────┘

Request (Anthropic → Gemini):

system → systemInstruction.parts[].text
messages[] → contents[] (role mapping: user/model)
tools[] → tools.functionDeclarations[] (schema conversion)
tool_choice → functionCallingConfig
max_tokens → generationConfig.maxOutputTokens
Tool results resolved via shadow store (tool_use_id → function name)

Response (Gemini → Anthropic):

candidates[].content.parts[] → content[] (text / tool_use blocks)
usageMetadata → usage (input_tokens / output_tokens)
finishReason → stop_reason (STOP→end_turn, MAX_TOKENS→max_tokens, SAFETY→refusal)
Blocked prompts → refusal with safety filter reason

Shadow Store

The shadow store (GeminiShadowStore) maintains per-session state to solve two problems:

Tool result name resolution — Gemini's functionResponse requires the function name, but Anthropic's tool_result only provides tool_use_id. The shadow store maps IDs to names.
Thought signature replay — Preserves assistant turn snapshots for multi-turn context without bloating the main proxy flow.

Bounded LRU cache prevents unbounded memory growth (max 200 sessions, 64 turns each).

URL Handling

Dual-mode URL normalization:

Structured (official Gemini endpoints): Normalized to canonical v1beta/models/{model}:generateContent format, preventing path duplication.
Opaque (custom relays): Used as-is with only query parameters appended, supporting arbitrary proxy setups.

Testing

40+ new unit tests covering:

URL normalization (structured and opaque endpoints)
Schema conversion (simple parameters and full JSON Schema)
Streaming conversion (text deltas, function calls, CRLF delimiters)
Shadow store lifecycle (record, read, eviction, clear)
Request/response transformation (messages, tools, tool results, blocked prompts)
Tool argument rectification from schema hints

Commits

abbf43ea — refactor(proxy): extract take_sse_block helper with CRLF delimiter support
3ce8245f — feat(proxy): add Gemini Native URL builder and full-URL resolver
8c09d8c7 — feat(proxy): add Gemini Native schema, shadow store, transform, and streaming
8603d4a3 — feat(proxy): wire Gemini Native format into proxy core and Claude adapter
70886d98 — feat(ui): add Gemini Native provider preset and api format option
8f300f45 — feat(proxy): add Gemini Native tool argument rectification
7cf554c0 — feat(proxy): update Gemini streaming and transformation logic

…pport Replace inline `buffer.find("\n\n")` SSE splitting logic across streaming, streaming_responses, response_handler, and response_processor with a shared `take_sse_block` function that handles both `\n\n` and `\r\n\r\n` delimiters.

Introduce gemini_url module that normalizes legacy Gemini/OpenAI-compatible base URLs into canonical models/*:generateContent endpoints. Supports both structured Gemini URLs (auto-normalized) and opaque relay URLs (pass-through with query params only).

…treaming - gemini_schema: Gemini generateContent request/response type definitions - gemini_shadow: session-scoped shadow store for thinking signature and tool-call state replay across streaming chunks - transform_gemini: bidirectional Anthropic Messages ↔ Gemini Native request/response conversion with thinking block and tool-use support - streaming_gemini: Gemini SSE → Anthropic SSE streaming adapter with incremental thinking/text/tool_use delta emission

…pter Integrate gemini_native api_format throughout the proxy pipeline: - ClaudeAdapter: detect Gemini provider type, Google/GoogleOAuth auth strategies, and suppress Anthropic-specific headers for Gemini targets - Forwarder: Gemini URL resolution, shadow store threading, endpoint rewriting to models/*:generateContent with stream/non-stream variants - Handlers: route Gemini streaming through streaming_gemini adapter and non-streaming through transform_gemini converter - Server/State: add GeminiShadowStore to shared ProxyState - StreamCheck: support gemini_native health check with proper auth headers

- Add gemini_native to ClaudeApiFormat type and ProviderMeta.apiFormat - Add "Gemini Native" provider preset with default Google AI endpoints - Show Gemini-specific endpoint hints and full-URL mode guidance - Add gemini_native option to API format selector in ClaudeFormFields - Add i18n strings for zh/en/ja

yovinchen · 2026-04-06T04:42:43Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7cf554c011

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-06T04:47:41Z

src-tauri/src/proxy/providers/transform_gemini.rs

+        .iter()
+        .filter(|message| message.get("role").and_then(|value| value.as_str()) == Some("assistant"))
+        .count();
+    let shadow_start_index = total_assistant_messages.saturating_sub(shadow_turns.len());


Align shadow replay to latest assistant turns

When a request contains fewer assistant messages than are cached in shadow_turns (for example after client-side history truncation), this offset calculation makes replay start from the oldest cached turn, so later assistant messages get mismatched snapshots. That can inject stale functionCall IDs/thought signatures into the reconstructed context and then cause downstream tool_result name resolution to fail. Use the tail of shadow_turns (latest N turns) when mapping cached turns to current assistant messages.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-04-06T04:47:41Z

src-tauri/src/proxy/gemini_url.rs

+    let path = path.trim_end_matches('/');
+    path.contains("/v1beta/models/")
+        || path.contains("/v1/models/")
+        || path.contains("/models/")


Narrow Gemini full-URL normalization matching

This heuristic classifies any full URL containing /models/ as a structured Gemini endpoint and rewrites it, which breaks opaque relay URLs that happen to include that segment (for example custom routes like /custom/models/invoke). In full-URL mode those relay paths are expected to be preserved, so this broad match can route requests to a non-existent rewritten endpoint. Restrict normalization detection to canonical Gemini patterns instead of a generic /models/ substring.

Useful? React with 👍 / 👎.

farion1231 · 2026-04-06T09:07:23Z

Hi, I owe you an apology for the long silence. As the sole maintainer of this project, I've been overwhelmed and fell behind on reviews — but that's on me, not on you. Your contribution matters and I will review this PR. If it needs a rebase, just let me know or I can handle it. Thank you for your time and patience.

# Conflicts: # src-tauri/src/proxy/providers/claude.rs # src-tauri/src/proxy/sse.rs # src-tauri/src/services/stream_check.rs

Resolve conflict in claudeProviderPresets.ts: keep both Gemini Native and Shengsuanyun presets.

Restore .unwrap_or(&provider.id) fallback for cache_key to match main branch behavior. Only gemini_native related changes should be in this branch.

yovinchen · 2026-04-11T11:27:23Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9bbc8bb789

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-11T11:33:15Z

src-tauri/src/proxy/providers/streaming_gemini.rs

+                    let text = String::from_utf8_lossy(&bytes);
+                    buffer.push_str(&text);


Preserve UTF-8 boundaries when buffering Gemini SSE

create_anthropic_sse_stream_from_gemini decodes each incoming byte chunk with String::from_utf8_lossy before appending to the SSE buffer. When a multibyte character is split across network chunks (common in streamed Chinese/Japanese text), lossy decoding inserts replacement characters and permanently corrupts content (and can distort downstream JSON payload text). The other SSE converters already avoid this by using append_utf8_safe with a remainder buffer; this path should do the same to prevent user-visible text corruption.

Useful? React with 👍 / 👎.

yovinchen added 7 commits April 6, 2026 11:09

feat(proxy): add Gemini Native tool argument rectification

8f300f4

feat(proxy): update Gemini streaming and transformation logic

7cf554c

chatgpt-codex-connector bot reviewed Apr 6, 2026

View reviewed changes

yovinchen added 4 commits April 6, 2026 23:42

fix(proxy): align shadow turns to tail on client history truncation

44ac926

Merge branch 'main' into feat/gemini-proxy-integration

92668f6

# Conflicts: # src-tauri/src/proxy/providers/claude.rs # src-tauri/src/proxy/sse.rs # src-tauri/src/services/stream_check.rs

Merge branch 'main' into feat/gemini-proxy-integration

d52a811

Resolve conflict in claudeProviderPresets.ts: keep both Gemini Native and Shengsuanyun presets.

fix: revert unrelated cache_key change in claude proxy transform

9bbc8bb

Restore .unwrap_or(&provider.id) fallback for cache_key to match main branch behavior. Only gemini_native related changes should be in this branch.

chatgpt-codex-connector bot reviewed Apr 11, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(proxy): Gemini Native API proxy integration#1918

feat(proxy): Gemini Native API proxy integration#1918
yovinchen wants to merge 11 commits intomainfrom
feat/gemini-proxy-integration

yovinchen commented Apr 6, 2026

Uh oh!

yovinchen commented Apr 6, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

chatgpt-codex-connector bot Apr 6, 2026

Uh oh!

chatgpt-codex-connector bot Apr 6, 2026

Uh oh!

farion1231 commented Apr 6, 2026

Uh oh!

yovinchen commented Apr 11, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

chatgpt-codex-connector bot Apr 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		let text = String::from_utf8_lossy(&bytes);
		buffer.push_str(&text);

Uh oh!

Conversation

yovinchen commented Apr 6, 2026

Summary

Motivation

Changes

New Modules (Backend)

Modified Modules (Backend)

Frontend

Architecture

Transformation Pipeline

Shadow Store

URL Handling

Testing

Commits

Uh oh!

yovinchen commented Apr 6, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Apr 6, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot Apr 6, 2026

Choose a reason for hiding this comment

Uh oh!

farion1231 commented Apr 6, 2026

Uh oh!

yovinchen commented Apr 11, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Apr 11, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants