Pixy: https://pixydesignapp.com
Docs: https://pixydesignapp.com/docs

# Create a project
One project is one website. It owns your edits and carries the three credentials — public key, browser approval, agent key.

A project is one website, or one codebase, if the same code serves localhost, staging and production. Everything you design flows through it: the script tag names it, the edits belong to it, and your agent pulls from it.

A new account has no projects. You create every one of them.

## Creating one

On [/my](https://pixydesignapp.com/my), type a name in the toolbar and press **New project**.

Call it whatever you call the site, `Acme Site`, `docs.acme.com`, `Blog`, up to 40 characters. The address comes from the name: `Acme Site` lives at `/my/projects/acme-site`. Free holds one project; a paid account holds as many as you like.

The card appears straight away with its setup open, the [script tag](https://pixydesignapp.com/docs/install) is the next step.

## The three credentials

One project, three strings, and they are deliberately not interchangeable.

- **The project key** (`px_…`) is public. It goes in the script tag, in your source, on production. All it does is name the project, on its own it opens nothing and shows nobody anything.
- **The browser approval** (`pxt_…`) is what actually grants editing. It is minted when you [approve a browser](https://pixydesignapp.com/docs/authorize), bound to that browser, that site and you, and stored in the site's own local storage. It never appears in your source.
- **The agent key** (`px_sk_…`) is private. It is what [your agent](https://pixydesignapp.com/docs/agents) authenticates with to pull edits and report back. It never appears on the card, and you never carry it by hand: the agent is handed one directly, by the one-time invitation inside the **Connect an agent** instructions.

That split is why the tag is safe on a live site. A visitor who reads your HTML gets the project's name and nothing else.

## Agents

Under the tasks, the card lists every agent holding a key on the project, the name it introduced itself with, when it joined, and when it last pulled. An agent that has been quiet for weeks says so, which is what makes the next part safe.

**Revoke** ends that one agent. Its next call is refused; everything it already built stays, and no other agent notices. It rejoins like it arrived: a new invitation.

An agent that is mid-build has no Revoke, the row says what it is building instead. The task is locked to that agent, and pulling its key out from under it would leave the lock with nobody able to finish or release it. Wait for it to report back, or release the task from the panel first; the Revoke returns either way.

## Regenerate

**Regenerate key** on the card is the panic button. Revoke for everything at once. It issues a new project key, revokes every agent key and every browser approval, and drops any page currently connected.

Everything the project owns, its pages, its edits, its tasks, survives untouched. What breaks is every deployed script tag: they still carry the old key. Copy the fresh tag from the card and redeploy, then approve your browser once more. Agents rejoin by fresh invitation, **Connect an agent** in the editor mints one.

## Remove

**Remove** deletes the project. Its script tag goes inert wherever it is still installed.