Download
Mini-Agent is a minimal yet production-minded demo project that shows how to build a serious command-line AI agent around the MiniMax-M2 model. It is designed both as a reference implementation and as a usable agent, demonstrating a full execution loop that includes planning, tool calls, and iterative refinement. The project exposes an Anthropic-compatible API interface and fully supports interleaved thinking, letting the agent alternate between reasoning steps and tool invocations during long, complex tasks. It includes a basic toolset for file-system operations and shell commands, plus integrations with MCP tools such as web search and knowledge graph access. Mini-Agent also comes with “Claude Skills”-style predefined skills for tasks like document processing, design work, and testing, packaged as reusable behaviors that can be invoked by the agent as needed.
Features
- Minimal but full-featured CLI agent that showcases best practices for building with MiniMax-M2
- Anthropic-compatible API usage with support for interleaved thinking during multi-step tasks
- End-to-end agent execution loop with tools for file-system access and shell command execution
- Persistent memory via a Session Note Tool to retain context and decisions across sessions
- Built-in skills for documents, design, testing, and development plus native MCP tool integration for search and knowledge graph access
- Clean, test-backed Python codebase with quick-start and development modes for rapid customization
Project Samples
