Quick start

By the end of this page you have one machine paired, one project with Lea as its lead, and a first conversation that turns into items on the plan. It takes about ten minutes. The examples use Dictly, an example dictation app.

You need:

• a Mac or a Linux machine where you already use the Claude Code CLI, the Codex CLI, or both, set up the way you like (for Codex, your ChatGPT plan through the Codex CLI’s login; for Claude Code, your own setup: an API key, or a sign-in Anthropic’s terms allow for this use);
• a local clone of the repo you want to work on, with an origin remote;
• git, and gh if workers should open pull requests.

1. Sign in. Open app.minga.run and create an account with your email and a password, or sign in with a passkey. If you are not in a workspace yet, Minga asks you to create one.

2. Install the runner on the machine that should do the work:

   curl -fsSL https://app.minga.run/install.sh | sh

   The script needs no sudo. It puts the minga binary in ~/.local/bin, checks its SHA-256 sum and, where an OpenSSL with Ed25519 is available, the release signature. It never edits your shell profile; if ~/.local/bin is not on your PATH, it prints the line to add.

3. Pair it with your workspace:

   minga runner login

   The runner prints a link and a code. Open the link (on macOS it opens by itself), check that the code matches, choose the workspace under Add it to and approve. The code expires after 10 minutes. This works over SSH too: open the link on any device where you are signed in.

4. Keep it running. Install the runner as a user service so it starts at login and restarts after a crash:

   minga runner install

   On macOS this is a LaunchAgent, on Linux a systemd --user unit. To try it out first, minga runner start runs it in the background without a service.

5. Check your CLI setups. On its first start the runner adds a profile for each CLI home it finds (~/.claude and ~/.codex). Check them:

   minga profile check

   Each profile should show as signed in, with the models it offers. If a profile is missing or you use a second CLI home, see Pair a runner and add profiles.

6. Register your repo so the runner knows where your clone lives:

   minga repo add ~/code/dictly

   The runner reads the origin remote and checks the default branch, read access and push access. If you have opened the repo in Claude Code or Codex on this machine before, the runner may already have found it; registering it makes the choice explicit.

7. Create a project. In the app, choose New project. Give it a name (“Dictly”) and pick the repo, then Create project. The lead is called Lea by default, and she runs on your runner. To change that, or to paste a brief you already have, open More options first: Where Lea runs offers your runners, or In the cloud if your workspace has an Anthropic or OpenAI API key. See Choose where Lea runs.

8. Answer Lea’s opening question. The project opens in Chat, where Lea asks what you are building. Describe the goal in a few sentences, for example: “Dictly 2.4 should get new users to their first dictation within a minute. Due November 15.” Lea asks what she needs to know, then writes it down.

What you see when it works

• The chip in the project bar reads Lea · working while she thinks, and free when she is done.
• Under each of Lea’s replies, a Saved line lists what the reply changed, with chips such as DIC-130 that link to the records, and Proposed with the cards it opened (P-3). A brief edit she proposes shows up as a card in Needs you, because the brief is yours.
• Plan shows a milestone with items under it. A new milestone waits in Next while you look it over. Moving it to Now is the go-ahead, and from then on workers start on the items whose checks are clear.
• Team shows each worker (W-12 · Welcome screen · my-mac) while it runs.

Runs use your own logins

Workers run the CLI with your git and gh logins and the profile’s access mode. The default, Auto, uses the CLIs’ own auto modes. Read Access modes and Data and security before you let work run unattended.

Next

• Your first project with Lea shows how a first conversation usually goes.
• Connect Claude Code and Codex over MCP lets your own sessions talk to Minga.