Release v2.1.0: Kimi API JSON Mode and Web Search Support

Release v2.1.0: Kimi API JSON Mode and Web Search Support

🎉 What's New

✨ Kimi API Advanced Features

This release adds full support for Kimi (Moonshot AI) API's advanced features, bringing the total Kimi API support to 100% for core functionality!

1️⃣ JSON Mode Support

• Structured Output: Force Kimi to output valid, parseable JSON
• Use Cases: Data extraction, structured responses, API integration
• Implementation: Automatic response_format parameter passthrough
• Example:

  {
    "response_format": {"type": "json_object"},
    "messages": [
      {"role": "system", "content": "Output JSON: {\"name\": \"...\", \"age\": ...}"},
      {"role": "user", "content": "我叫张三，今年25岁"}
    ]
  }

2️⃣ Web Search Support

• Built-in Search: Use Kimi's native $web_search function
• Zero Configuration: No need to implement search/crawl functions
• Automatic Handling: Proxy intercepts and returns arguments to Kimi
• Example:

  {
    "tools": [{
      "type": "builtin_function",
      "function": {"name": "$web_search"}
    }]
  }

🐛 Bug Fixes

Debug Log Cleanup

• Removed excessive debug logging from OpenAI message construction
• Cleaner console output for better debugging experience

Tool Definition Filtering

• Removed edit-file from tool definitions sent to AI models
• Prevents wasted round trips when AI tries to use unsupported tools
• AI now directly uses str-replace-editor for file editing

File Editing Interruption Fix

• Fixed issue where file editing operations would cause conversation to be interrupted
• Conversation now continues properly after intercepted tool execution
• Correct stop_reason handling (3 instead of 1)

📊 Kimi API Support Matrix

Feature	Support	Notes
Streaming	✅ 100%	SSE format, [DONE] terminator, streaming tool_calls
Tool Calls	✅ 100%	Parallel calls, index field, arguments accumulation
Thinking Mode	✅ 100%	reasoning_content field support
JSON Mode	✅ 100%	response_format parameter support
Web Search	✅ 100%	$web_search builtin_function support
Partial Mode	⚠️ 50%	Theoretical support, depends on Augment
File Interface	❌ 0%	Not in proxy scope
Token Calculation	❌ 0%	Not in proxy scope
Balance Query	❌ 0%	Not in proxy scope

Core Functionality: ✅ 100%
Advanced Features: ✅ 83.3% (5/6)

🔧 Technical Details

Modified Files

• src/providers/openai.ts: Added responseFormat parameter support
• src/tools.ts: Added builtin_function type support and $web_search interception
• src/config.ts: Added Kimi provider configuration
• package.json: Updated version to 2.1.0
• README.md: Added JSON Mode and Web Search feature descriptions
• CHANGELOG.md: Documented all changes

Implementation Highlights

JSON Mode:

// Automatically extracts and passes response_format to Kimi API
const responseFormat = augmentReq.response_format || undefined;
if (responseFormat) {
    requestBody.response_format = responseFormat;
}

Web Search:

// Intercepts $web_search and returns arguments as-is for Kimi to execute
if (toolName === '$web_search') {
    return {
        toolName,
        input,
        intercepted: true,
        result: input  // Return arguments unchanged
    };
}

📦 Installation

Download augment-proxy-manager-2.1.0.vsix and install via:

• VSCode: Extensions → Install from VSIX
• Command line: code --install-extension augment-proxy-manager-2.1.0.vsix

🚀 Upgrade Notes

No breaking changes. Simply update to v2.1.0 to get all new features.

📝 Full Changelog

See CHANGELOG.md for complete version history.

---

Previous Release: v2.0.0
Repository: LegnaOS/VSC-Augment-Proxy-Manager