How it works

Your work app is three pieces on your own laptop: a folder of files, a browser tab at http://localhost:9950 that shows them, and Claude in the Code tab of the Claude desktop app, which changes them. This page shows how the three fit together. The pages below go into each part.

The three pieces

Piece	What it is	What you do with it
The folder	Your app: its data files, its wiki pages, its code	Nothing directly. You open it in the Code tab, and Claude reads and edits it
The browser tab	The app's screen, served from your laptop at `localhost:9950`	Read your modules, tick tasks, add and edit rows
The Code tab	Claude, opened on the folder	Ask for changes, new modules, imports, fixes and explanations

localhost means "this computer". The address works only on your laptop, and only while the app's server is running. Claude starts that server for you and keeps it running on port 9950. If the tab ever says it cannot connect, ask Claude: "Is my app running? Start it if not."

What runs where

Everything runs on your laptop. There is no hosting, no sign-in and no database.

Part	Where it lives	Who changes it
Data files	`app/data/`, one per module	You on screen (datatable rows only) and Claude
Wiki pages and About	`app/wiki/` and `app/data/about.md`	Claude
Module files	`app/src/modules/`, one per module	Claude
Blocks and other code	`app/src/`	Claude, when content alone cannot do it
The dev server	A program on your laptop	Claude starts and restarts it

Claude asks for your approval before it runs a command or changes files, depending on your settings. You stay in charge of every change.

Nothing leaves your laptop

The app itself sends nothing anywhere. Its data is plain files in your folder, and the screen reads them through a small server on your own machine.

Two things do go beyond the laptop, and you control both:

Your conversation with Claude. What you type, and the files Claude reads to answer you, go to Claude like any chat does.
Connectors, when you add them. A connector lets Claude read from a tool such as SharePoint or Jira, with your own permissions. See Agents and connectors.

The app runs on your laptop, for you. Nothing is published.

One change, two ways

Sam, a customer success manager at Brightlane, has their app open. Here is the same kind of change made from each side.

From the screen. In Tasks they tick "Send Tallow & Co the renewal proposal". The browser sends the change to the server on their laptop. The server writes the row's status, done, into app/data/tasks.json.

From Claude. In the Code tab they type: "Add Daniel Reyes's colleague Eva Strand to People: Head of operations at Harbourline Logistics, met at the rollout call on 6 October." Claude adds a row to app/data/people.json. Within about five seconds, the People page in their browser reloads by itself and shows the new row.

Both changes end up in the same files. That is the whole design: one set of data files, two writers, one screen.

The pages in this section

Page	Read it to understand
Data files	Where your data lives, and what happens when you and Claude both write
Modules and blocks	How the sidebar is made, and how a new module needs no new code
Fields and views	Columns, field kinds, tabs, and the table and task views
The wiki and About	How pages are written and why they use Markdoc
Agents and connectors	How Claude reaches the tools where your work lives