Summary

Maki is an AI coding agent optimized for minimal context token usage and a superior user experience. It targets developers seeking a fast, lightweight, and transparent tool for code interaction, offering significant savings in API costs and improved performance.

How It Works

Maki achieves context efficiency via its index tool (using tree-sitter for precise code parsing) and code_execution tool (a sandboxed monty interpreter for data processing without context inflation). Task delegation supports variable AI model strengths. System prompts are concise, and optional rtk integration further reduces bash output tokens.

Quick Start & Requirements

Installation is recommended via a reviewed curl script (https://maki.sh/install.sh). Alternatives include cargo (main branch), Nix (nix run github:tontinton/maki), or pre-built binaries. Maki supports Anthropic, OpenAI, Copilot, Ollama, Mistral, Z.AI, and Synthetic providers (API keys required), plus custom providers via scripts.

Highlighted Details

• Performance: ratatui-based TUI offers 60 FPS and low memory usage; splash screen uses SIMD.
• Security: Unique bash command permission parsing via tree-sitter and SSRF protection for web requests.
• Transparency: Full visibility into subagents and agent actions; no hidden information.
• Extensibility: Supports custom AI providers, UI-less JSON streaming, and long-term memory management.

Maintenance & Community

The project notes >90% of its code is AI-generated, guided by humans. Specific community links or maintainer details are absent from the provided text.

Licensing & Compatibility

No license information is specified in the README, posing a significant adoption blocker, especially for commercial use.

Limitations & Caveats

The absence of a stated license is the primary limitation. The AI-generated nature of the codebase implies potential quality variations. Chat history rewind is available, but code rewind is not yet implemented.