Skip to content
Fredrin
Documentation

Get started

Quickstart

Install Fredrin, connect a repository, run one ticket, and merge the pull request it opens. The whole loop, once, in about fifteen minutes.

Fredrin is a desktop app. Everything below happens on your machine, against a real repository, using the AI coding subscription you already pay for.

Before you start

You need two things: a git repository you can push to, and an AI coding CLI that is already signed in on this machine (Claude Code is the default).

1. Install and open the app

Download Fredrin for your platform and open it. On first launch it runs a setup preflight: a one-pass check of every tool a Worker needs on this machine, with a real detected version per row.

Fix anything the checklist flags before moving on. On Windows a missing required row holds the app closed rather than failing later at the moment of use.

2. Connect a repository

Point Fredrin at a repository. It clones it if you do not have it locally, and the board you land on is that project's backlog.

A project is one repository. If you work across several, add them all: each gets its own board, its own tickets, and its own running Workers.

3. Write your first ticket

Create a ticket describing one change you want. Keep it to a single unit of work, the kind of thing you would open one pull request for.

A good first ticket is small, real, and verifiable. Fix a bug you already know about. Add a missing test. Rename a confusing function everywhere it appears. Avoid anything that needs a design decision you have not made yet.

The description is the prompt, so write it the way you would brief a competent engineer who does not know this codebase: what you want, where it lives if you know, and how you will tell it worked. Writing a good ticket goes deeper.

4. Run it

Press Run. Fredrin does four things:

  1. Creates a branch for the ticket.
  2. Creates a real git worktree for that branch, an isolated checkout of your repo.
  3. Starts an AI agent session, its Worker, in that worktree.
  4. Streams the work back onto the card as it happens.

You can watch it in the chat tab, or open the terminal tab and see the raw session. You do not have to watch. That is the point.

Try Plan first if you are unsure. The Plan / Build toggle runs the same Worker in planning-only mode: it writes no code, asks up to three questions only a human can answer, and ends with a saved plan you can read before flipping it to Build on the same ticket.

5. Review the diff and merge

When the Worker finishes, the ticket lands in Review. Read the diff. That is the job.

If it is right, ship it. Fredrin opens the pull request, and its CI status flows back onto the card. When the PR merges the ticket auto-completes and moves itself to Completed.

If it is not right, send it back with a comment. The Worker picks up where it left off in the same session, with the same context, in the same worktree.

Now run three at once

The loop above is the whole product, and running it once does not show you why it exists. Create three tickets and run them together.

Each gets its own branch, its own worktree, and its own Worker. They do not collide, and they do not wait for each other. Your job becomes reviewing three diffs instead of babysitting one terminal.

That is the shift Fredrin is built around: throughput comes from parallelism, not from a faster agent.

Next steps