What humanize is

humanize runs flows: Python files that drive one or more coding agents in a loop, and write down everything they did so it can be read back afterwards.

It does not talk to a model provider. It drives the coding agent CLI you already have — claude, codex, kimi, pi, opencode, mimo — logged in the way you already log in. There is no API key to give it and none for it to hold.

What it is for

A single conversation with a coding agent is a chat window. Most work is not that shape:

You want	humanize's answer
The agent to keep going until the job is done, starting fresh each time	A Ralph loop
One agent building and another reading its work	A two-agent flow
Two hundred files fixed at once	A fan-out
The same run twice, unattended, in CI	hmz exec and CI
To know what a nine-hour run actually did	hmz collect
One CLI driven as two accounts at once	Providers
The work to land in a container, or on the build box	Containers, remote execution

Two ways in

A terminal interface. hmz with nothing after it opens a transcript, an editor, and a status line. Pick a flow with /flow, say what you want done, and watch it. There is no hmz tui: one way in is one way in.

A command line. hmz exec -f <flow> -a <agent> "<task>" runs the same flows with nobody watching — which is what a script, a cron entry or a CI job wants.

Both write the same cycle, and both are read back the same way.

What to read

• Never used it. Installation → Getting started → Concepts.
• Want a feature. Features has a page each — /afk, skills, goals, containers, providers, tracing — and each says how to reach it from the prompt, the command line and Python.
• Automating something. The tutorials go from your first run to publishing a flowverse, in order.
• Looking something up. CLI and TUI are complete.
• Changing humanize itself. Architecture.

Before you point one at a repository you care about, read Security. humanize runs every agent with permission prompts disabled, and there is no setting that turns them back on.