Agentic AI in Practice: Building AI Agents with Chinese Models
AI Agents are the dominant application pattern of 2026. This article covers how to build AI Agents with planning, tool calling, and memory capabilities using ChinaWHAPI model combinations.
AI Agent Core Components
A complete AI Agent includes: planning module (decides next steps), tool calling (executes external operations), memory module (saves context), and execution loop (reasons until completion).
Planning Module
Use DeepSeek R1 for complex task planning and decomposition — it's strong at multi-step thinking and logical reasoning.
{"model":"deepseek-r1","messages":[{"role":"user","content":"The user wants to build an e-commerce website. Please break this down into specific development steps"}]}Tool Calling
Chinese models (like DeepSeek V4 series, Qwen3.6 Plus) support the Function Calling tool protocol. Define tools like get_weather, search_db, etc., and the model autonomously chooses which to call.
Agent Framework Selection
LangChain, LlamaIndex, AutoGen, and CrewAI all support OpenAI-compatible interfaces and work directly with ChinaWHAPI.
- LangChain: most mature agent framework
- CrewAI: great for multi-agent collaboration
- AutoGen: Microsoft's conversational agent