Skip to main content
Cogno does the work for you — but first, a one-time setup connects it to your accounts and your computer. It takes about fifteen minutes. By the end, your first task is implemented and waiting for you as a pull request on GitHub, and Cogno runs on its own from there.

​
Before you start

You’ll need three accounts. If you already have them, just use those.
  • Cogno — sign up at app.cogno.studio.
  • Claude Code or Codex — a claude.ai or OpenAI account for the coding agent that writes the work. You sign its CLI in on your own machine in step 3; Cogno never handles those credentials.
  • GitHub — github.com, where Cogno delivers its work back to you as pull requests.
Everything else — the desktop app and its setup — happens in the steps below.

​
1. Create your Cogno account

1

Open the app

Go to app.cogno.studio in your browser.
2

Sign up

Create an account with Sign up with Google, Sign up with Apple, or email. For email registration, use a password with at least 6 characters and select Create account. Then enter the 6-digit confirmation code from your inbox on the Check your email screen. Continuing is your agreement to the Terms of Service and Privacy Policy linked below Create account — there is nothing to tick. In the desktop app those two links open in your default browser.
3

Set up your profile

The Welcome to Cogno screen asks for your name (a photo is optional), then drops you into your workspace.

​
2. Install the desktop app

Your work runs on your own computer, and the desktop app is what puts it there — it installs the Engine, the always-on runtime that runs agent work on this machine.
1

Download it

In the browser app, open your avatar menu and select Download desktop app — or go straight to cogno.studio/desktop. Choose your operating system.
2

Install and sign in

Install it, then sign in with your Cogno account. Signing in enrolls this computer as a machine — no separate setup wizard.
3

Check the machine

Open Settings → Machines. Under This machine, the Engine panel reads Enrolled and Connected when everything is in place. If a tool is missing — the agent CLI, git, or the GitHub CLI — the page says so and shows the fix.
On the same page, Startup & background → Open Cogno at login puts Cogno in the menu bar from the moment you sign in, so work can run without the window open. On macOS the close button only puts the window away — Quit Cogno in the menu bar is what ends it.

​
3. Sign in to your coding agent

Cogno does not write the code itself — it hands the work to Claude Code or Codex, running on this machine as you. So the agent’s own CLI has to be installed and signed in before any task can run. Cogno never sees or stores those credentials; the CLI holds its own. You only need one of the two. A machine with either one signed in can run work.
1

Install the CLI

If it isn’t already on this machine, install the one you have an account for:macOS and Linux
Windows (PowerShell)
Both are self-contained binaries in ~/.local/bin — neither needs Node.js. If a machine has no curl, both also publish npm packages: @anthropic-ai/claude-code and @openai/codex.
2

Sign in

Run the agent once in a terminal and complete its login in the browser:
This is a one-time step per machine, and it is yours to do — a browser sign-in cannot be automated, and Cogno deliberately does not ask for your Anthropic or OpenAI credentials.
3

Confirm Cogno can see it

Back in Settings → Machines, the Engine panel’s readiness line should stop naming the agent. Cogno re-checks on each check-in by asking the CLI itself — claude auth status or codex login status — so the line clears on its own once you are signed in.
The agent CLI is signed out on the readiness line means work cannot run on this machine and will wait rather than fail. Signing the CLI back in clears it — a token that expires or is revoked later shows up the same way. See Read the Readiness line.
Sign the GitHub CLI in on the same machine while you are in the terminal — gh auth login. Without it a run still writes code, but the readiness line reads The GitHub CLI is signed out, so no pull request will be opened.

​
4. Create a project and connect a folder

A project is where tasks live, and it names the folder on your machine the work runs in — its environment.
1

Create the project

In the sidebar, select + beside Projects. The menu offers the folders on this machine — Recents, On this Mac, and Open folder… — plus the repositories of a connected GitHub. Select the folder your code is already in: the project is created, named after what that folder holds, and its page opens. There is no form to fill in.
2

Check where it runs

A folder picked from the menu is connected as your environment at the same time, so there is nothing left to do. Choosing under GitHub instead looks for a folder on this machine holding those repositories, and asks where to clone them when there is none.If the project’s page says no machine can run it yet, its Environments section is where you connect one — Choose folder for a checkout you already have, or a Location to clone into.
An environment without GitHub can only edit a local folder — connecting GitHub is what lets runs open pull requests, and what fills the menu’s GitHub list. Connect it from Integrations.

​
5. Run your first task

1

Create work

Select Work in the header. Give it a title and describe what you want. No real task in mind? Try:
Build a Notion-style homepage in Next.js.
2

Send it off

Under How to proceed, select Watch it get built (recommended) — the task is created and a run starts on your machine immediately. Plan it first drafts a plan without building; Later drops it in the backlog. The environment is picked automatically.
3

Make sure a machine is running

If your machine is asleep, open the task’s assignee picker: your own row carries Start my machine.
Tasks also arrive on their own — suggested from Slack, meetings, and your other tools — see Connect integrations.

​
6. Review the result

The task’s conversation reads Agent is working… while the run is in flight and fills with the agent’s own messages as it goes. When the run finishes, the agent’s final message tells you what it did — there’s nothing else to click through. If the run opened a pull request, the task shows the delivery — repository, branch, and PR number — and you review and merge on GitHub. Want to pick the work up yourself instead? See Take over a task in your own terminal.

​
You’re all set

Your accounts are connected, the Engine is running on your machine, and your first task just came back as a pull request. From here, Cogno runs on its own — the more tools you connect, the more tasks arrive and resolve without you lifting a finger.

​
What’s next

  • Connect integrations — bring in Slack, GitHub, and more so tasks suggest themselves.
  • Run an agent on a task — how runs start, where they execute, and how to steer them.
  • Concepts — machines, environments, and sessions, and how the loop fits together.