feat: Add model routing configuration support

[papperrollinggery]

Summary / 概述

This PR adds Model Routing feature to cc-switch, allowing a single Provider to route different Claude models to different backends based on the requested model name.

本 PR 为 cc-switch 添加了模型路由功能，允许单个 Provider 根据请求的模型名称自动路由到不同的后端服务。

Problem / 问题

Before (之前)	After (之后)
Single apiFormat per Provider	Multiple routing rules per Provider
All models routed to same backend	Model-specific routing
Multiple Providers needed for different formats	One Provider handles all

Example / 示例:

// Before: Only one apiFormat
{ "meta": { "apiFormat": "anthropic" } }

// After: Multiple routes per Provider
{ "meta": {
    "apiFormat": "anthropic",
    "modelRoutingConfig": {
      "enabled": true,
      "routes": [
        { "sourceModel": "Claude Opus", "target": { "apiFormat": "openai_chat", "modelName": "gemini-2.5-pro" } },
        { "sourceModel": "Claude Sonnet", "target": { "apiFormat": "openai_chat", "modelName": "gemini-2.5-flash" } }
      ]
    }
  }
}

---

Related Issue / 关联 Issue

Fixes #

---

Technical Details / 技术细节

Architecture / 架构

Claude Code → cc-switch → NewAPI (localhost:3000) → AI Platforms
                      ↓ [routing by model]
              Claude Opus   → gemini-2.5-pro (OpenAI Chat)
              Claude Sonnet → gemini-2.5-flash (OpenAI Chat)
              Claude Haiku  → MiniMax-Text-01 (Anthropic)

Routing Flow / 路由流程

1. Extract model field from request body (case-insensitive)
2. Look up matching sourceModel in modelRoutingConfig.routes
3. Override base_url, api_format, and model_name based on target config
4. Fall back to default config if no route matches

Files Changed / 修改的文件

Backend (Rust) - 后端:

• src-tauri/src/provider.rs - Added ModelRoutingConfig, ModelRoute, RouteTarget, RouteFallback structs
• src-tauri/src/proxy/providers/claude.rs - Added get_claude_api_format_with_model(), get_routed_base_url()
• src-tauri/src/proxy/forwarder.rs - Integrated routing in 3 locations (lines 764, 870, 1383)
• src-tauri/src/proxy/providers/mod.rs - Exported new functions

Frontend (TypeScript) - 前端:

• src/components/providers/forms/ModelRoutingConfigPanel.tsx - NEW - UI panel component (426 lines)
• src/components/providers/forms/hooks/useModelRoutingConfig.ts - NEW - State management hook (90 lines)
• src/components/providers/forms/ProviderAdvancedConfig.tsx - Integrated ModelRoutingConfigPanel
• src/components/providers/forms/ProviderForm.tsx - State and persistence management

i18n Files / 国际化

• src/i18n/locales/zh.json - 中文翻译 (+18 lines)
• src/i18n/locales/en.json - English translation (+18 lines)

---

Testing / 测试

Build Verification / 构建验证

Test	Command	Status
TypeScript check	pnpm typecheck	✅ Pass
Format check	pnpm format:check	✅ Pass
Rust check	cargo check	✅ Pass
Rust clippy	cargo clippy	✅ Pass
Production build	pnpm tauri build	✅ Pass

Manual Testing / 手动测试

1. Open cc-switch → Add/Edit Provider
2. Set Base URL to NewAPI endpoint (e.g., http://localhost:3000)
3. Expand "Advanced Configuration"
4. Enable "Model Routing Configuration"
5. Add routing rules:
   • Source Model: Claude Opus → Target: gemini-2.5-pro (OpenAI Chat)
   • Source Model: Claude Sonnet → Target: gemini-2.5-flash (OpenAI Chat)
   • Source Model: Claude Haiku → Target: MiniMax-Text-01 (Anthropic)
6. Save and test with different Claude models

Edge Cases / 边界情况

Case	Expected Behavior
No matching route	Use fallback.apiFormat if configured
No fallback configured	Use default Provider apiFormat
Routing disabled	Use default Provider behavior
Case mismatch	Case-insensitive matching (e.g., "Claude Opus" = "claude opus")

---

Breaking Changes / 破坏性变更

None / 无 - This feature is additive and backward-compatible.

Existing Providers without modelRoutingConfig work unchanged. The feature only activates when modelRoutingConfig.enabled = true.

---

Checklist / 检查清单

• pnpm typecheck passes
• pnpm format:check passes
• cargo clippy passes
• Updated i18n files
• Backend compiles successfully
• Frontend compiles successfully
• Application builds and runs

---

Commit / 提交

feat(provider): add model routing for single Provider

- Support routing different Claude models to different backends
- Add ModelRoutingConfig data structures to provider.rs
- Implement get_claude_api_format_with_model() and get_routed_base_url()
- Add ModelRoutingConfigPanel UI component
- Add useModelRoutingConfig state management hook
- Integrate routing logic in forwarder.rs
- Update zh.json and en.json translations

---

Notes / 备注

1. API Key Management: This implementation assumes NewAPI handles API key injection. Direct routing to external APIs requires additional key management.

2. Claude Adapter Only: Current implementation is for Claude adapter only. Other adapters (Gemini, MiniMax, etc.) can be extended similarly.

3. Security: All routing happens on the client side. Ensure the NewAPI endpoint is trusted and properly secured.

[farion1231]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 78b08fa38e

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]

Stop leaking strings when selecting routed API format

get_claude_api_format_with_model now allocates and leaks memory on every matched route by returning Box::leak(...into_boxed_str()). Because this path runs per request, a busy long-lived proxy process will accumulate unreclaimed heap memory over time. This should return a non-leaking owned value (or map to known static literals) instead of manufacturing 'static strings at runtime.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Rewrite request model when a routing rule matches

The routing integration in forward only replaces base_url (and separately API format), but it never updates the outgoing request model to RouteTarget.model_name. As a result, routes intended to map Claude model IDs to different backend model IDs still send the original Claude model upstream, which can cause wrong-model execution or upstream rejection for non-Claude backends. The matched route’s target model needs to be applied to the request body before sending.

Useful? React with 👍 / 👎.