Skip to content
Lattis Lattis

Introduction

Lattis is a desktop workspace for running AI coding agents across multiple branches. Each agent can work in its own branch and worktree, while Lattis provides a shared view of their sessions and status.

Lattis can also be used from the web when you are away from your desktop. The web app connects to a running desktop so you can check branches, start sessions, attach to a terminal, and handle a few branch actions. The work still runs on your machine. See Remote Control for the current scope and limitations.

It’s built around three jobs:

  1. Run several agents. Launch Sawyer (built in) plus Claude, Codex, Pi, and OpenCode side by side, each in its own git worktree.
  2. See what needs attention. Agents and branches show working, blocked, and waiting states, along with pull-request, review, and CI status in the app.
  3. See usage and cost information. Requests are attributed and priced when the provider reports enough information, so you can compare usage by project, model, and conversation.

Run several agents

Section titled “Run several agents”

Add a project and open a branch. Lattis starts an agent in a fresh worktree with its own terminal. You can open multiple branches and run several agents in parallel, including two agents on the same task for comparison. Review their output as it becomes available.

Monitor agent and branch status

Section titled “Monitor agent and branch status”

The sidebar shows stoplights: a green dot is working, amber is waiting on you, and red indicates an error. Each branch also shows its live pull request and CI state, so you can review, merge, push, and pull from the app when those integrations are configured.

Review changes

Section titled “Review changes”

When several agents or branches are in progress, Lattis also provides tools for reviewing them. Pending Reviews lists pull requests your team asked you to read, and opens a review workspace with a guided diff, draft comment threads, and an optional read-only agent pass. Your verdict posts to GitHub as anchored line comments. The reviewer picker can rank people who contributed to the changed code.

Usage and cost information

Section titled “Usage and cost information”

Because agent traffic flows through one local endpoint, Lattis can attribute usage and cost information. The usage and cost view breaks available data down by project, model, and conversation. The optional web app can aggregate available data from connected desktop apps.

Where to next

Section titled “Where to next”