Kowalski 2.2: download it, answer three questions, send in a horde

Three releases went out this week, and together they finish the job we set ourselves in the summer: kowalski should be something you download and use, not something you build.

2.0 — Out of the Box

One binary with the whole app inside. Start kowalski, a browser opens, and a Setup screen asks three questions: which AI (a free local model through Ollama, or any OpenAI-compatible service with your key), which folder the agents may work in, and, if you want spreadsheets, your tableski account. It ships with the Spreadsheet analyst, which turns plain questions into a report workbook without letting the AI compute a single number.

2.1 — Standing Orders

• Download and run. Pre-built binaries for macOS (Apple Silicon and Intel) and Linux (x86_64 and ARM). The installer fetches the right one and checks it.
• Two hordes that work while you do something else. The Morning brief reads the pages you follow and writes one page, at 7:00 on weekdays if you switch it on. The Folder watcher turns each document you drop into its inbox into a note with the facts and the deadlines.
• Safe by default. Steps that run a command or change files stop and ask. Workflows imported from elsewhere run isolated, with their schedules off. A kowalski reachable from your network always asks for its token.

2.2 — Commando

A new look you can actually read: penguin black, snow white and one signal red, in light and dark. The screens are named for what you do (Hordes, Chat, Build, Setup), and the Hordes screen shows each run as a row of steps, then puts the answers on the page when it is done. The docs got the same treatment, starting with a new architecture page.

Get it

curl -fsSL https://raw.githubusercontent.com/yarenty/kowalski/main/install.sh | bash
kowalski

Everything is open source under MIT at github.com/yarenty/kowalski; the full list of changes is in the changelog. If you want the long version of why the numbers can be trusted, read Ask your spreadsheets.