pi-zen
A quiet Pi TUI on one switch: backgroundless messages, a rail editor, compact tool calls, and restrained reasoning
Package details
Install pi-zen from npm and Pi will load the resources declared by the package manifest.
$ pi install npm:pi-zen- Package
pi-zen- Version
0.2.0- Published
- Sep 11, 2026
- Downloads
- 416/mo · 220/wk
- Author
- roshvan
- License
- MIT
- Types
- extension
- Size
- 1.4 MB
- Dependencies
- 0 dependencies · 2 peers
Pi manifest JSON
{
"extensions": [
"./extensions/zen.ts"
]
}Security note
Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.
README
pi-zen
I built pi-zen because I wanted a minimal interface that made it easier to stay present in an agent session. I did not want to keep scrolling through long stretches of context to find the few details I needed; I wanted a clean way to follow along, ask questions, and steer and ride the loop.
For real knowledge work, I like being an active participant. I want to understand what is happening between the AI and me so we can be good partners in making decisions. pi-zen gives me that space: clear responses, visible progress, and the useful parts of the session in a calm, visually appealing interface.
pi-zen is a presentation extension for the Pi terminal interface. It removes visual clutter while preserving conversations, tool activity, reasoning, and session data. It adds compact tool summaries and diffs, a minimal editor rail, a quieter startup, a calmer working indicator, and a live reading of where the current request's time is going, without changing tool execution or model prompts.
Showcase
Light mode

Dark mode

Coding cat demo

While a request runs
Above the editor, one dim line says where the time is going and updates as the request runs:
7.2s thinking · 3.6s writing · ↑6.5k ↓5.8k
Thinking is the time the model spends reasoning, including the silence before it streams anything; writing is the time it spends producing text and tool calls. Time a tool spends running belongs to neither and is left out. The token counts follow Pi's footer: tokens sent, excluding cache reads and writes, and tokens produced, reasoning included. Both cover the whole request, however many turns it takes. A half of the work that took no measurable time is left out, so a model that does not stream its reasoning simply reads as writing. The final reading stays until the next request starts.
Quick start
Requires Pi 0.84.3 or newer.
Install the extension from npm:
pi install npm:pi-zen
You can also install the latest version directly from GitHub with pi install git:github.com/Roshvan/pi-zen. Start Pi as usual, then run /zen to toggle Zen mode.
Development
You will need Node.js 22.19 or newer and pnpm. Clone the repository, install the dependencies, and start Pi with the local extension:
git clone https://github.com/Roshvan/pi-zen.git
cd pi-zen
pnpm install
pnpm dev
Before submitting a change, run pnpm check and pnpm pack:check.
Issues and contributions
Issues and pull requests are welcome. If you have an idea, find a bug, or want to improve something, feel free to open an issue or create a pull request. I am happy to look it over.