Threads Overview

A Thread is an isolated workspace for working on a project. Each thread is:

• A full clone of the project’s repository in its own directory
• On its own git branch, derived from the thread title
• Paired with an independent chat session with the AI agent
• Configured with its own skill and model preferences

Threads are the core unit of work in Tarsk. When you ask the AI agent to make changes, it operates inside that thread’s clone — never touching any other thread or the original project directory.

Why Threads Exist

Traditional AI coding tools operate on a single shared copy of your codebase. This creates a problem: if you want to try two different approaches to the same feature, or work on two issues simultaneously, edits conflict.

Threads solve this by giving each line of work its own isolated environment. You can:

• Run the agent on feature A in thread 1 while reviewing feature B in thread 2
• Try two completely different approaches to a refactor and cherry-pick the better result
• Keep a “clean” thread for reference while experimenting in another
• Work on a bug fix in one thread while a new feature is being developed in another

Each thread has its own git branch, so merging the result back is a standard git merge or pull request.

Thread Storage

Threads are stored at:

~/Library/Application Support/Tarsk/data/{threadId}/

This is a real directory on disk containing a full git clone. You can open it in any IDE, run commands in it, or inspect it with any git tool. Changes made by the AI agent are real filesystem writes — not diffs or patches.

Thread State

Field	Description
Title	Human-readable name (e.g. “Add login page”)
Branch	Auto-generated from the title (e.g. add-login-page)
Current Conversation	The active conversation ID within this thread
Enabled/Disabled Skills	Per-thread skill overrides
Status	active or deleted

Next Steps

• Working with Threads — create, switch, open in IDE, and delete
• Dev Server — run and preview your app inside a thread