feat: Add support for Anthropic SDK (Claude) and multi-provider switching

[SmartisanNaive]

This PR introduces native support for the Anthropic (Claude) SDK to MiroFish!

While the project already supports various models via the OpenAI-compatible interface, integrating the official Anthropic SDK allows us to better leverage the capabilities of models like Claude 4 Sonnet. It also ensures compatibility with providers that follow the Anthropic protocol (e.g., Zhipu AI's GLM-4.7 endpoint).

🛠️ Key Changes

1. Refactored LLMClient (backend/app/utils/llm_client.py):

   • Introduced an LLM_PROVIDER mechanism, allowing seamless switching between openai (default) and anthropic.
   • JSON Mode Adaptation: Since the Claude SDK lacks a native json_object mode, I implemented a fallback strategy within the client using System Prompt injection and Markdown parsing. This ensures a consistent chat_json experience for the upper layers.

2. Configuration Enhancements (backend/app/config.py & .env.example):

   • Added new config options: ANTHROPIC_API_KEY, ANTHROPIC_BASE_URL, etc.
   • Organized .env.example to clearly distinguish between provider configurations.

3. Documentation:

   • Added backend/LLM_README.md with detailed instructions on how to configure and switch providers.

🧪 How to Test

1. Update dependencies:

   cd backend
   uv sync  # or uv pip install -r requirements.txt

2. Update .env to switch to Anthropic:

   LLM_PROVIDER=anthropic
   ANTHROPIC_API_KEY=your_key
   # ANTHROPIC_BASE_URL=... (optional, for custom endpoints)

3. Start the backend service and trigger a prediction task to verify that the logs output correctly.

Verified locally using Zhipu GLM-4.7 (Anthropic protocol) - communication works as expected.

Ready for review! Thanks 🚀

[666ghj]

I appreciate it, but considering that supporting an additional API format would bring unnecessary workload to the subsequent update plans, I will keep this PR to provide ideas for everyone, but it will not be merged before the release of version 1.0.

[SmartisanNaive]

Thanks for reading , this is a great project.Your project has brought a lot of inspiration to our researchers working on multi-agent applications, thanks again(●'◡'●)

…

---Original--- From: ***@***.***> Date: Sun, Jan 25, 2026 00:20 AM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [666ghj/MiroFish] feat: Add support for Anthropic SDK (Claude)and multi-provider switching (PR #38) 666ghj left a comment (666ghj/MiroFish#38) I appreciate it, but considering that supporting an additional API format would bring unnecessary workload to the subsequent update plans, I will keep this PR to provide ideas for everyone, but it will not be merged before the release of version 1.0. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>