Welcome to the Tarsk blog

Tarsk is a personal agent that builds things. It runs on your machine, works inside real Git repositories, and gives each task its own isolated copy of the code. This blog is where we explain how that works and why we built it.

What you can expect

We will write about the parts of Tarsk that are hard to see from the outside:

• Threads. How isolated working copies let you run several agents in parallel without collisions.
• Agents and tools. What the model sees, how tools return results, and how we keep long runs from drowning in context.
• Providers and models. Working with any provider, and what we learn when we compare them on real tasks.
• Release notes. What changed, and what to try first.

How to follow along

New posts appear on this page. If you want to run Tarsk while you read, install the CLI and open it in your browser:

npx tarsk

The documentation covers installation, projects, threads, and settings. The skills and MCP directory lists extensions you can add to a thread.