Add Anthropic Messages API support for Zhipu GLM coding plan

[adisusilayasa]

Summary

This PR adds support for the Anthropic Messages API format, enabling PicoClaw to work with Zhipu's GLM coding plan and other Anthropic-compatible APIs.

Changes

• Auto-detect Anthropic-style API based on apiBase path containing /anthropic
• Split Chat() into two methods:
  • chatOpenAI() - handles OpenAI-style chat completions
  • chatAnthropic() - handles Anthropic Messages API format
• Implement proper Anthropic Messages API format:
  • Content as array of blocks (not single string)
  • Separate tool_use and text content blocks
  • Correct headers (x-api-key, anthropic-version)
  • Support both /v1/messages and /messages endpoints
• Add parseAnthropicResponse() for parsing Anthropic-style responses

Why This Matters

This enables users to use:

• Zhipu GLM coding plan (https://api.z.ai/api/anthropic/v1)
• Model: glm-4.7 (and other GLM models)
• Other Anthropic-compatible APIs

Testing

Tested successfully with:

Provider: zhipu
Model: glm-4.7
API Base: https://api.z.ai/api/anthropic/v1

Example test:

User: "2+2"
Assistant: "2 + 2 = 4" ✅

Configuration Example

{
  "agents": {
    "defaults": {
      "provider": "zhipu",
      "model": "glm-4.7"
    }
  },
  "providers": {
    "zhipu": {
      "api_key": "your-api-key",
      "api_base": "https://api.z.ai/api/anthropic/v1"
    }
  }
}

🤖 Generated with Claude Code

[yinwm]

We make a provider related refactor in #283 , could you please change the code into new provider way.

Another, Z.AI can use zhipu provider in model_list, using OpenAI-style, is Anthropic-style really needed?

@adisusilayasa

[adisusilayasa]

Thank you for the feedback! After deep research into Zhipu GLM coding plan official documentation (docs.z.ai), I've confirmed that:

1. Zhipu GLM coding plan's PRIMARY API format is OpenAI-style

   • Primary endpoint: https://api.z.ai/api/coding/paas/v4/chat/completions
   • Official documentation states: "Z.AI provides interfaces compatible with OpenAI API"
   • Provides OpenAI SDK examples for Python, Node.js, and Java

2. Anthropic-style endpoint is SECONDARY

   • Secondary endpoint: https://api.z.ai/api/anthropic
   • Purpose: Only for Claude Code integration, not general API usage
   • Documentation explicitly states: "The Anthropic endpoint is specifically for integration with Claude Code and similar tools, not for general API usage"

Conclusion

Zhipu GLM coding plan works perfectly with the existing zhipu/ protocol using OpenAI-style API. Therefore, Anthropic-style API support is not needed.

What Was Changed

I've refactored this PR to use the new protocol-based provider architecture from PR #492:

Documentation Changes Only (No Code Modifications)

1. Added Zhipu GLM configuration to config/config.example.json

   {
     "model_name": "glm-4.7",
     "model": "zhipu/glm-4.7",
     "api_key": "YOUR_ZHIPU_API_KEY",
     "api_base": "https://open.bigmodel.cn/api/paas/v4",
     "_comment": "For Zhipu GLM coding plan, use api_base: https://api.z.ai/api/paas/v4"
   }

2. Updated README.md (English)

   • Added documentation explaining how to configure Zhipu GLM coding plan with custom api_base
   • Showed example with https://api.z.ai/api/paas/v4 endpoint

3. Updated README.zh.md (Chinese)

   • Same documentation translated to Chinese

How Users Configure Zhipu GLM Coding Plan

With the new architecture, users can simply configure it in model_list:

{
  "model_list": [
    {
      "model_name": "glm-4.7-coding",
      "model": "zhipu/glm-4.7",
      "api_key": "your-zhipu-key",
      "api_base": "https://api.z.ai/api/coding/paas/v4"
    }
  ],
  "agents": {
    "defaults": {
      "model": "glm-4.7-coding"
    }
  }
}

This uses the existing, proven zhipu/ protocol with OpenAI-style API - zero code changes needed!

Addresses Reviewer Concerns

✅ "Could you please change the code into new provider way?"

• Done: Uses new protocol-based model_list configuration from PR feat(config): refactor provider architecture to protocol-based model_list #492
• No auto-detection or Anthropic-specific code needed

✅ "Z.AI can use zhipu provider in model_list, using OpenAI-style, is Anthropic-style really needed?"

• Answer: NO, Anthropic-style is not needed
• Zhipu GLM coding plan's primary and officially documented API format is OpenAI-style
• The existing zhipu/ protocol implementation already supports it with just a custom api_base

The PR now contains only documentation updates showing how to use Zhipu GLM coding plan with the new protocol-based architecture. This is a simpler, more maintainable approach that avoids the compatibility issues encountered with the Anthropic-style implementation.