KOWALSKI

Squad // open source · MIT · macOS & Linux

Hordes of small AI agents that do real work on your machine.

Kowalski is one binary. Answer three setup questions, hand your spreadsheets, pages and documents to a horde, and get files back you can open: a report workbook, a morning brief, a note with deadlines. Your model, your files, your machine.

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

Get it on GitHubHow it is built

Three questions to the Spreadsheet analyst on a local 7B model. The minutes it spends thinking are cut.

Horde 01

Spreadsheet analyst

Ask questions in plain words. Get report.xlsx, one sheet per question. The model writes SQL; every number comes from the query engine.

Horde 02

Morning brief

The pages you check every morning become one page: a top pick and what matters from each, with links. Runs at 7:00 on weekdays if you switch it on.

Horde 03

Folder watcher

Drop an invoice or a contract into the inbox. Get a note: what it is, the key facts, and what to do by when.

How it works

One process. Engines compute, models write.

The UI, the horde runner and the agent core run in one binary on 127.0.0.1. Tools come in over MCP, starting with tableski for SQL over spreadsheets. Steps that must be right — fetching, profiling tables, running SQL, building workbooks — use no model at all. Commands wait for your approval.

Read the architecture →

Kowalski architecture: one binary, the horde runner, and the tools it talks to

From the blog

Latest

All posts →