OleksandrChekhovskyi/hax

What it solves

It provides a lightweight, terminal-native coding agent for developers who want a fast, resource-efficient tool that doesn't take over the terminal or require heavy IDE integrations. It specifically addresses the need for an agent that works well with local models and respects Unix philosophy.

How it works

Written in C as a single native binary, hax connects to various LLM providers (including local ones like llama.cpp and Ollama) to act as a coding assistant. It uses a minimalist approach, avoiding complex plugin runtimes or IDE panels, and instead relies on subprocesses for composition and plain-text configuration files.

Who it’s for

Developers who live in the terminal, prefer running local models, need to audit the exact data sent to and from the model, or operate in resource-constrained environments.

Highlights

• Minimalist footprint: A small C binary that starts instantly and uses very little memory.
• Local-first: Auto-discovers capabilities when connected to llama.cpp.
• Terminal-native: Supports streaming Markdown and live tool output while preserving native scrollback.
• Inspectable: Includes a transcript view (Ctrl+T) to see the exact wire protocol and model interactions.
• Broad compatibility: Supports OpenAI, Anthropic, OpenRouter, llama.cpp, Ollama, and custom endpoints.