mini-cc is a lightweight, multi-language AI programming agent and educational project. It dissects and replicates core architectures of large-scale AI agents, focusing on tool-use, sandbox execution, and memory compression. Targeting developers and researchers, it offers a functional, declarative terminal UI and a plugin ecosystem for practical learning and building advanced AI agent systems.

How It Works

The agent uses a pure functional loop for tool invocation. Key components include an AgentTool system for isolated subprocess command execution and an .ai_memory engine for context compression (data stripping, history truncation) to prevent token overflow. The Model Context Protocol (MCP) enables secure, cross-process tool extension via plugins, facilitating safe network and system operation integration.

Quick Start & Requirements

Available in TypeScript, Python, Go, and Rust. Installation varies:

• TypeScript: npm install -g @you-want/mini-cc
• Python: pip install you-want-mini-cc
• Go: go install github.com/you-want/mini-cc/go/cmd/mini-cc@latest (Go >= 1.21)
• Rust: cargo install --git https://github.com/you-want/mini-cc.git --bin minicc (Rust 1.70+) Source builds available. Initial setup requires API key configuration (e.g., OPENAI_API_KEY) via an interactive wizard. A React (Ink) terminal UI provides a rich interactive experience.

Highlighted Details

• Multi-Model Support: Integrates with Anthropic, OpenAI, and compatible APIs (Qwen, DeepSeek, Kimi).
• Declarative Terminal UI: Dynamic, stream-based UI built with React (Ink).
• MCP Plugin Ecosystem: Enables secure, seamless extension of agent capabilities.
• Secure Sandbox: Bash sandbox with command stripping and high-risk command interception.
• Easter Eggs: Includes /buddy companion and /voice simulated voice chat.

Maintenance & Community

No specific details regarding maintainers, community channels (Discord/Slack), or active development signals were found in the provided README.

Licensing & Compatibility

Released under the MIT License, permitting free use, modification, and distribution, including for commercial purposes, with standard attribution.

Limitations & Caveats

Advanced architectural concepts, like cross-platform screen capture or cloud cluster inference, are presented as 'architecture演练' (demos) in src/architecture-mocks and lack actual implementation. The project's primary focus is educational, serving as a teaching tool for agent architecture replication.