Rich PoC & simulator/fake connector

[noirbizarre]

Hi 👋🏼

Since Pycon FR 2025, I've been using PyInfra a lot (thanks @wowi42 for the presentation!)

I wanted to contribute but didn't have time before. But now I have.

Rich

I always thought using rich could really improve the user experience, and I saw #785, so I gave it a try, coupled with Cyclopts (which gives typing, completion... and a few other benefits over Click).

Also, I did not (yet) address the suggested verbose mode from this comment (eg. fact/operation level details...). I think this is a very interesting feature, but it requires a base rich integration to be working, and it's a lot more work (not a single contribution IMHO).

Showcase

Here's some demo, with the current state and the rich version (click to see full size).

Inventory

Current	Rich

Python Webapp success

Current	Rich

Python Webapp with errors

Current	Rich

Help

Current	Rich

Simulator / Fake connector

To make it easier for me to test and compare, I have written a small simulator. Basically a fake programmable connector:

• answer successfully by default
• use host data fake_answers dict for explicit answers (success or failure, using stdout or stderr...)
• response times are randomized and configurable

It has been incredibly useful (and actually allows me to easily test other things), so I'm thinking maybe it would be useful to someone else.

PR(s) or not PR(s)?

The current work is on the my refactor/rich-cyclopts branch. The first commit is the simulator, and then the rich+cyclopts refactoring.

Before going further, I need to know if you would be interested in those contributions. Only the rich refactoring? Only the fake connector? Both (as separate PRs obviously)?
What should be focused on before contributing ?

Note

Yes, I am using AI for development and I used AI to help on this contribution. I prefer to be transparent and don't want this to be an issue. (However this post is entirely manually written :) )

Any feedback is welcome!

[wowi42]

This is so good. The before/after shots don't lie: Rich version is next-level.

Fake connector stealing the show. Testing without infra? Yes please, all day.

Two PRs: smart. Simulator first for the quick win, then we can nerd out over CLI libraries.

PS: I think that was FOSDEM 2026. Either way, glad it brought you here.

Keep it up. This rocks.

[noirbizarre]

So let's go !! 🚀

Here's the first PR, the low hanging fruit: the fake connector.

The https://github.com/noirbizarre/pyinfra/tree/refactor/rich-cyclopts branch has already been rebased on top of it showing how it can be used.

PS: You are right, it was FOSDEM 2026! 😅

[noirbizarre]

Now the rich part: #1894

[wowi42]

Moved over from PR #1894 (now closed) so the details live here:

Migrates the pyinfra CLI from Click to Cyclopts for argument parsing and adopts Rich for all human-facing output, while keeping the existing command syntax and behaviour.

Highlights:

• Cyclopts-based CLI with grouped --help, colourised usage, shell completion, and a machine-readable --json mode (stdout stays pure JSON; all human output goes to stderr).
• Rich debug-inventory: hosts rendered in a zebra-striped table with data shown as key: value lines (nested values as JSON), type-coloured to match the JSON highlighter.
• Live hierarchical progress tree during deploys, with per-host/per-operation nodes and informative error handling that attributes failures to the right host and operation.

Output is routed through a small pluggable layer in pyinfra.api.output, decoupled from Rich/Click so the core stays testable. Existing behaviour, exit codes and command syntax are preserved; a few intentional UX changes (usage-error exit code, --version output, confirm-prompt re-prompting) are documented in the commits.

The implementation remains available on the closed PR branch (#1894) if we decide to pick it up.