humanfia/humanize

The agent flow system for token maxxing.

humanize (hmz) – an agent‑flow orchestrator for AI‑powered coding

What it is

  • humanize (invoked as hmz) is a command‑line tool that lets you run coding agents (e.g., Claude Code, DeepSeek, Kimi, or any model reachable through litellm) as part of a configurable flow.
  • Its tagline, “The agent flow system for token maxxing,” hints that the tool is meant to help you chain together prompts and actions while keeping an eye on token usage.

Key features (as described in the README)

  • Multiple back‑ends – support for DeepSeek ([dsh]), Kimi ([kimi]), and generic litellm models. You can install one, some, or all of them with the extra‑dependency syntax hmz[all].
  • CLI and web UI – run hmz in a terminal for a text‑only interface, or launch hmz web to open a browser‑based UI that shows every “epic” (a flow) on the machine.
  • One‑line flow execution – the exec sub‑command lets you run a specific flow directly from the shell, e.g.:
    hmz exec -f chat -a assistant=claude/claude-opus-5:high "What does this repository do?"
    
  • Desktop app – on macOS you can download a signed DMG that simply wraps the hmz web UI in a native window.
  • Automatic updates – hmz update pulls the latest release while preserving the extras you originally installed.

Installation

  1. Prerequisites – Python 3.12+, the uv package manager, and an account with a supported coding‑agent CLI (Claude Code, Codex, etc.).
  2. Install the core tool:
    uv tool install hmz
    
  3. Add the back‑ends you need (choose one, some, or all):
    uv tool install 'hmz[all]'   # installs DeepSeek, Kimi, and litellm support
    
  4. Update later with hmz update or run a one‑off version via uvx:
    uvx hmz@latest
    uvx --from 'hmz[all]' hmz web
    

Basic usage

  • hmz – starts the terminal UI inside the current repository.
  • hmz web – opens the web UI showing all flows on the machine.
  • hmz exec … – runs a specific flow directly from the command line (see example above).

Documentation & help

Community & contribution

  • Pull requests are welcomed; contribution guidelines are in CONTRIBUTING.md.
  • Bugs, feature requests, and security reports have dedicated forms and policies.
  • The project follows the Apache‑2.0 license and a community Code of Conduct.

Bottom line humanize is a lightweight orchestrator that lets developers plug in various LLM‑backed coding agents, define flows, and run them either in a terminal or a browser UI, with a focus on managing token consumption. It is positioned for developers who already use AI coding assistants and want a unified way to chain their actions.

Related

  • Project
  • Project
  • Project
  • Project