DIY Your AI Agent

中文文档 | deepwiki

This project is part of the whitejoce/AI-Agent-Toolkit stack, focusing on the Agent layer.
Full architecture: RAG (Enterprise Knowledge Base) → Agent → Tool Runtime (Hot-reloadable MCP Tools Platform)

🔥 Project Overview

Translated by GPT-5.5

This is a minimal runnable AI Agent example. It does not depend on frameworks such as LangChain or LangGraph. The project is designed to be lightweight, readable, and easy to modify, making it suitable for learning the basic structure of an Agent and its tool-calling mechanism.

Mini Agent

The MVP example in this repository lives in mini_agent/.

• Usage guide: mini_agent/README_EN.md
• Entry point: mini_agent/agent.py
• Tool definitions: mini_agent/tools.py

Project Structure

.
├── requirements.txt    # Python dependencies
├── mini_agent/
│   ├── agent.py        # Agent loop: model calls, tool dispatch, terminal interaction
│   ├── tools.py        # Tool schemas and execution handlers
│   ├── README_*.md     # Documentation
│   └── .env.example    # Environment variable example
├── img/demo.png        # Demo screenshot
├── README_CN.md        # Chinese README
├── README.md           # English README
└── LICENSE

Roadmap

Keep Mini Agent as the learning and testing base. A more complete version can gradually add the following modules:

• ToolRegistry: manage built-in tools, third-party tools, and MCP tools in one place.
• ApprovalPolicy: ask for user confirmation before high-risk actions such as writing files or running commands.
• ContextManager: manage short-term context, conversation compression, and token budgets.
• Memory: store long-term memory, user preferences, and project-level context.

Safety Note

This repository is better suited for learning the basic structure of Agents. For daily use, prefer mature community-maintained projects.

💡 See Also

1. Ask the Friendly AI

• LLM APIs
  • OpenAI Responses API, Chat Completions API
  • Anthropic Messages API
• MCP protocol, Skills, and progressive context
• Agentic AI
  • Why are prompts becoming less central?
  • Observability and orchestration: logs, tool-call records, error tracing, and performance monitoring
  • Human-in-the-loop?
• Scoring and evaluation
  • Artificial Analysis, Deep SWE benchmark
  • Context engineering vs Harness engineering
  • Function Calling, Structured Outputs
• Multi-Agent: Agent SDK
  • ADK: A2A protocol, agent.json
  • Design ideas and implementation details of frameworks such as LangGraph and LangChain
• Measuring LLM output quality
  • Metrics: accuracy, completeness, length, and path

2. Context Management Trade-offs

Recommended: Claude Code's animated context-window demo.

• Short-term memory: intelligently select and preserve the most relevant information in the current conversation
  • What is the dumb zone?
  • Context compression: summarize previous conversation history to save tokens while preserving continuity
• Long-term memory: remember user preferences, conversation history, and project background to improve personalization and continuity
  • AGENT.md, CLAUDE.md: global and project-level context files
  • Memory mechanism: persist command history, user preferences, and related information
• External knowledge bases: Retrieval-Augmented Generation (RAG)

3. Explore Different Harness Design Ideas

What is a Harness? Why is it central to Agent design?

• OpenClaw, Hermes Agent, OpenHuman, Pi
• Coding Agent: Cursor, Codex, Claude Code, OpenCode
  • Hard Core: understand how Pi works

📜 License

This project is released under the MIT License.

🤝 Contributions

Issues and PRs are welcome.