Lightweight by design: a single native C binary with a small dependency set starts instantly and uses only few MBs of memory, leaving more RAM for local LLMs. Local models are first-class; hax auto-discovers the model and runtime capabilities without custom config for the default setup.

Respects your terminal: streaming Markdown and live tool output reflowed for display, preserving native scrollback. It only redraws the current streaming line or input area. Inspectable via transcript view (Ctrl+T). Supports OpenAI, Anthropic, Codex, OpenRouter, llama.cpp, and more.

Target audience: developers who live in the terminal, run local models, audit tool actions, or operate where resources are scarce. It deliberately omits MCP marketplaces and plugins; see docs/philosophy.md. Runs on Linux and macOS; on Windows under WSL. Install via Homebrew or download prebuilt static binary for x86_64 or aarch64. Build from source with `make`.

Quick start: run `hax` in your project directory for interactive REPL, use `-p "list TODOs"` for one-shot, or `-c` to continue latest session. Use `/provider` to see available providers. hax remembers your interactive selections.