Unstoppable Code is a desktop app for macOS and Windows that runs AI agents on your computer. You give each agent a task, it works in its own copy of your project, and you review what it did.
This article covers what you do with the app day to day. For the bigger picture of why it exists, read Why Choose Unstoppable Code? on our blog.
Run several tasks at once
Each task is a chat with an agent. In a Code project, every new task also gets its own worktree: a separate copy of your repository on its own branch, so one task's changes stay apart from another's.
Select New task at the top of the sidebar and describe the work. While that agent works, start another task. Each agent changes only its own worktree, so they do not overwrite each other.
Plan before the agent writes code
In a Code project, turn on Plan mode from the composer's + menu. The agent drafts a plan and waits for your review before it changes any code.
Open the plan and comment on any passage. A button such as Submit 3 comments sends your comments to the agent. When you are ready, the app asks Implement this plan? Choose Yes, implement this plan, or Run it as a pipeline.
Review the changes
When the agent finishes, open its changes in the diff view. Comment on any line and send the comments to the agent, the same way as plan comments. When the work looks right, select Create PR. The app commits and pushes what it needs to, then opens a pull request.
For a second opinion, run the bundled Code Review (Multi-Provider) pipeline. It reviews the changes with both Claude and GPT models and can post its findings as pull request comments.
Automate with Workflows
Workflows in the sidebar holds the work that runs on your behalf, in three tabs:
- Automations run a chat or a pipeline on a schedule, or once at a set time.
- Pipelines are saved series of steps, such as agents, scripts, and checks, that run as one job. Start one with Pipeline in a new task's + menu, from a plan, from a hook, or from an automation.
- Watchers wait for one event, such as a pipeline run ending, then send a message to a chat so it can continue. Agents usually create them for you.
To react to pull request events, such as failing checks, merge conflicts, or new review comments, add a hook in the project's settings under Hooks.
Check in from anywhere
With Remote Control, you can follow your agents and answer their questions from a browser or the mobile app. Your agents still run on your computer, so keep the desktop app open while you are away.
Use Work for tasks without code
Work in Unstoppable Code is for tasks that are not code, such as writing, research, and reports. A Work project holds files and chats, with no branches or worktrees. When you add a folder that is not a git repository, it becomes a Work project. When you create a new project, choose the Cowork type.
You can also switch the whole app to Work mode, which uses plain language and keeps developer controls one step away. Open Settings, go to General, and change Mode. The Work page shows what Work is built for.
What is free, and how agents run
Every app feature is free when you sign in. Agents need a model to run, and you can supply one in three ways:
- Your own subscription. Use your Claude, ChatGPT, or Cursor subscription to run Claude Code, Codex, or Cursor on a plan you already have.
- Local models. Run models on your own computer through Ollama.
- AI credits. Turn on 25 free AI credits a day in your account, and buy more when you need them. AI credits can also keep a chat going when you reach a Claude or Codex usage limit.
AI credits and usage limits explains how the two kinds of usage work together.
Next steps
- Connect your AI models so your agents have a model to run.
- Start your first task to try it on a real project.
- The Glossary defines the terms in this article.