Guides
One page per thing humanize does, each answering "how do I use this?". Every guide opens with a worked example you can paste, then explains the rest.
These are for looking things up. If you have not used humanize before, the tutorials teach it in order, starting with the Quickstart.
Start here
| Installation | Python, a backend, and what each one needs |
| Concepts | The ten words the rest of this uses |
| Security | Read this before pointing an agent at a repository you care about |
| Troubleshooting | When it goes wrong |
At the prompt
| Talking to a running turn | A line typed mid-turn goes into it, not after it |
| Many conversations at once | One transcript, tab between the agents that are working |
| Showing the working | /details: tool calls and thinking, or only what the agent says |
| The shape of a run | /status: who is working, who handed to whom, what it cost |
| Being away | /afk: whether an agent may stop and ask you something |
| Completion | What a half-typed line could become, under the editor |
| History | Everything typed here before, on ↑ and ↓ |
| Exporting a transcript | /export writes what is on screen, as it was written |
| What a project remembers | Reopening finds it set up the way you left it |
| Stopping | esc ends the flow; what that does to a turn |
Setting an agent up
| Efforts | How hard to think — and moving it while the flow runs |
| Permissions | Four rungs, from read-only to bypass |
| Skills | What an agent carries: its CLI's own, and the ones the flow brings |
| Goals | The backend's own goal feature: it decides when it is done |
| Questions | An agent stopping mid-turn to ask its user something |
| Answers in a shape | A turn that answers with a pydantic model instead of prose |
| Hooks | Python callables hung on the moments of a turn |
| Cost and rate | What has been spent, how fast, and how hard it is thinking |
| The person as an agent | You, driven by a flow like any other agent |
| Reporting | What humanize sends its developers, and how to say no |
Writing flows
| Writing a flow | The dozen lines that make a directory a flow |
| Loops | Ralph, stateful ralph, and the shapes a loop takes |
| Settings of its own | A pydantic model that becomes /config fields |
| Many turns at once | async def run, and awaiting several turns |
| A flow that calls a flow | Composition, and whose agents the inner one gets |
| Testing a flow | Checking the loop without spending a turn |
| Flowverses | A git repository of flows, offered by name |
Where the work lands
| Providers | One CLI, two accounts, at the same time |
| Containers | A container of the agent's own, up on the first turn |
| Remote execution | The agent here; its commands on the build box |
| Worktrees | One agent working in several directories at once |
Running it, and reading it back
| Unattended | hmz exec from a script, with nobody watching |
| humanize in CI | The same flows on a build machine |
| Tracing | The whole run as one timeline you can open in Perfetto |
| Picking a run up | A loop stopped on Thursday, carried on from where it stopped |
Looking for the exhaustive list of flags and keys instead? CLI and TUI.