Native terminal agent · written in Zig

An agent that runs where your shell does.

hmn is one native binary. Start it in a project directory and it reads, edits and runs things there, on your machine or inside a Linux container. Point it at any OpenAI-compatible provider.

Install Release v0.1.0

$ curl -fsSL https://h.hmbrd.dev/install.sh | sh

Verified on macOS with Apple Silicon. Other platforms have not been verified yet. You can read the script before you run it.

~/project

# save your token once
$ hmn login

# start a session in this directory
$ hmn

# the same, with no permission prompts
$ hmn --yolo

# carry on where you stopped
$ hmn session resume last

01What it does

Lives in the terminal

hmn opens an interactive session by default. For a single request that prints its answer and exits, use hmn ask "…".

Your provider, your choice

Connect to an OpenAI-compatible Chat Completions endpoint that supports tools. The hmbrd relay is the default; provider flags select another connection.

Save your token once

hmn login writes the token to a credential file inside your hmn profile. The default path is ~/.hmn/credentials.json, with permissions 0600.

Choose the permission mode

The default auto mode applies the agent’s permission policy. Use --permission-mode ask for approvals or --yolo for full access without permission prompts.

Sessions are kept

Sessions are stored in your profile and associated with a workspace. hmn sessions lists them, and hmn session resume last reopens the latest one.

Tools are built in

File editing, shell commands and network tools such as Nmap, DNS lookup and WHOIS are native tools with structured results, not text scraped from a terminal.

02Provider

Out of the box hmn talks to the hmbrd relay. Sign in once with a token from hmbrd.dev and it is ready.

Endpoint: https://api.hmbrd.dev/v1
Model: model-zero
Token: private file in your profile, mode 0600

Default provider

$ hmn login

Any other OpenAI-compatible provider

$ hmn --custom-provider NAME \
    --base-url https://example.com/v1 \
    --model MODEL

First save the token with hmn login --custom-provider NAME --base-url https://example.com/v1 --model MODEL. Run flags override the connection for one session; login saves it.

03Where it runs

The same agent runs in two places. Pick the host for everyday work in a repository. Pick the container when you want the agent and everything it runs kept away from the rest of your machine.

Host

Directly in your shell

The native binary, your files, your tools. This is what the installer sets up.

Container

A Linux environment

A Kali Linux arm64 image with the agent and its tools inside, started through OrbStack or Apple Container. It sees only the profile and the workspace you mount, and needs no privileged mode.

The image is about 3.2 GiB and downloads separately on first use. Run hmn --runtime orbstack --yolo or select --runtime apple. Specify the target in chat.

04Platforms

Platform	Status
macOS, Apple Silicon	Verified first
macOS, Intel	Not verified yet
Linux	Not verified yet, outside the container image
Windows	Not verified yet