Maki – the efficient coder (AI agent)
11 hours ago
- #Code Efficiency
- #AI Optimization
- #Developer Tools
- Parses 15 languages into skeletons to reduce token usage, saving 165 tokens per turn on reads.
- Features a sandboxed Python interpreter with async tools for efficient data processing without cluttering context.
- Includes subagents with varying strengths (weak, medium, strong) and access levels, optimized for different tasks.
- Automatically compacts history to manage long context by stripping unnecessary elements like images and thinking blocks.
- Offers a native binary with performance optimizations like SIMD and background syntax highlighting for smooth user experience.
- Displays real-time information like token count, cost, and model in the status bar, with multiple subagent chat windows.
- Supports bash command parsing with tree-sitter for security, allowing per-tool rules or bypass with --yolo.
- Provides long-term memory across sessions, plan mode for read-only access, and MCP server integration with 26 themes.
- Uses indexing to parse file structures and read only needed lines, improving efficiency by avoiding full file reads.
- Enables sandboxed code execution via async Python functions, where only print output enters the context for clarity.