hax

A minimalist, terminal-native coding agent written in C.

Key features

• Lightweight by design — A single native C binary with a small dependency set. Starts instantly and uses only a few MB of memory, leaving more RAM for local models.
• Local models are first-class — Start llama-server -m [model].gguf, then hax --provider llama.cpp, and hax auto-discovers the model and runtime capabilities. No custom provider config block is needed for the default setup.
• Respects your terminal — Streaming Markdown and live tool output, reflowed for display in the terminal. Only redraws the current streaming line or the input area, native scrollback is preserved. Does not take over or mess with your terminal.
• Inspectable — See exactly what was sent to the model and what it replied in a usable transcript view (Ctrl+T). Optionally collect a detailed wire protocol trace.
• Broad provider support — OpenAI (+compatible), Anthropic (+compatible), Codex (via a ChatGPT subscription), OpenRouter, OpenCode Zen/Go, DeepSeek, llama.cpp, Ollama, and custom endpoints.
• Well-behaved Unix tool — XDG paths, clean stdout in -p one-shot mode with resume hints on stderr, plain-text config and session files, composition via subprocesses instead of plugins.

Target audience

Developers who live in the terminal, run local models, audit what their tools do, package software for distros, or run agents where resources are scarce. If you want MCP marketplaces, a plugin runtime, IDE panels, or per-command permission prompts, other agents build exactly that — hax deliberately doesn't, and docs/philosophy.md explains each omission and the pattern that covers the need.

If "fancy new AI tech in an old-school minimalist package" sounds like your vibe, you might like this.

Install

hax runs on Linux, macOS, FreeBSD, and OpenBSD; on Windows, use it under WSL. The BSDs build from source only.

With Homebrew (macOS or Linux):

brew install oleksandrchekhovskyi/hax/hax

On Arch Linux, hax is in the AUR as hax.

Alternatively, prebuilt binaries for Linux (x86_64 and aarch64) and macOS (Apple Silicon) are available from the latest release: unpack the hax binary into any directory on your PATH. The Linux binaries are fully static, and the macOS one needs only libraries that ship with macOS. If macOS refuses to run a binary downloaded through a browser, clear its quarantine flag with xattr -d com.apple.quarantine hax.

From source

Building from source gives a binary linked against your system's shared libraries instead of the libraries the prebuilt binaries bundle:

git clone https://github.com/OleksandrChekhovskyi/hax.git
cd hax
scripts/install_deps.sh   # Debian/Ubuntu, Fedora, Arch, openSUSE, Alpine, macOS, FreeBSD, OpenBSD
make                      # the binary is now at ./build/hax
make install              # optional; may prompt for sudo

scripts/install_deps.sh installs the build dependencies — a C compiler, libcurl, jansson, meson, ninja, and pkg-config — plus fzf, which hax uses for @file completion when available. On other platforms, install those packages by hand and run make.

For hacking on hax, make symlink links the freshly built binary into ~/.local/bin so it stays on PATH across rebuilds. make tests additionally needs tmux, which drives the interactive scenarios (scripts/install_deps.sh tests installs it), and make lint needs clang-format and clang-tidy (scripts/install_deps.sh lint installs them).

The examples below use hax as if it is on PATH; after a plain build, use ./build/hax.

Connect a provider

The easiest first run is interactive: start hax, then use /provider to see available providers and choose a model. hax remembers interactive provider, model, and effort selections.

Provider	Setup
codex	Run /login inside hax.
openai	Set OPENAI_API_KEY.
anthropic	Set ANTHROPIC_API_KEY.
openrouter	Set OPENROUTER_API_KEY.
opencode-zen / opencode-go	Set OPENCODE_API_KEY.
deepseek	Set DEEPSEEK_API_KEY.
llama.cpp	Run llama-server.
ollama	Run ollama serve.
Compatible or custom endpoint	See docs/providers.md.

Quick start

Run hax from the project directory you want it to work in:

hax                         # interactive REPL
hax -p "list TODOs"         # run one prompt and print the final answer
printf "explain x" | hax -p # read the prompt from stdin
hax -c                      # continue the latest session for this directory
hax --resume                # pick a past session for this directory
hax --resume=ID -p "next"   # resume a specific session in one-shot mode

Run hax --help for CLI usage, or man hax where the man pages are installed. In the REPL, type /help for slash commands and shortcuts. See docs/usage.md for more detailed usage documentation.

Configuration

Configuration is optional. Interactive selections are remembered separately from your config file; CLI flags apply only to the current run, and environment variables are useful for shells and scripts. Resumed conversations restore their own provider, model, effort, and preset unless a CLI selection flag overrides them.

See docs/configuration.md for the file format, resolution order, presets, and the setting reference.

Extras

Optional companions that live outside the core binary, each a small Python CLI with a skill that tells hax when to use it:

• hax-web — web search and readable page fetching.
• hax-mcp — tools of MCP servers, with definitions entering the model's context only as a task needs them.

More docs

• docs/sessions.md — the session-file format and the hax --json stream.
• docs/debugging.md — trace/transcript logs, mock provider, and demo scripts.
• CONTRIBUTING.md — how to propose, prepare, and submit a change.

License

MIT.