humanizeOrchestrate, execute, and observe agent flows

Drive the coding agent CLIs you already have — one flow, many agents, one timeline you can read afterwards.

One flow, many agents

A flow is a Python file that says what each agent is asked, in what order, and when to stop. Ship it, run it by name, or run one of the flows humanize already offers.

Write one

Six coding agents, one interface

claude, codex, kimi, pi, opencode and mimo, driven through the CLI you already log into. humanize holds no API key and talks to no model provider itself.

How it fits

A prompt you can watch

A transcript per conversation, tab to step between the agents that are working, and a line typed mid-turn that goes into the turn rather than after it.

Talk to a running turn

Two accounts of one CLI

An agent may name the account it runs as, so one flow drives one CLI as a subscription and somebody else's endpoint at the same time.

Providers

Somewhere other than here

Give an agent a container of its own, or moor it to an ssh host so its commands land there while the process stays on this machine.

Remote execution

The whole run as a timeline

Every run is written down as it happens. Collect it into a Chrome trace and open it in Perfetto — one process per agent, one track per session, one slice per thing it did.

Tracing

Install it

pip install git+https://github.com/humanfia/humanize2.git

Run something

hmz

Opening the humanize interface, typing / to see the commands, and picking a flow

Recorded against a stand-in coding agent, in a container of its own — see Working on these docs.

Or without the interface, over the agents you name, one -a apiece:

hmz exec -f official/flame_chase \
    -a claude/claude-opus-4-8:high -a codex/gpt-5.6-sol:high "fix the build"

Then read the whole thing back:

hmz collect

New here? Getting started goes from nothing installed to a run you can read back, and Concepts defines the ten words the rest of this uses.