Profiles that stay signed in
Each profile is its own person, with its own sign-ins, cookies and tabs. Close Kulisa, open it tomorrow: everyone is still signed in, with the same tabs open.
Be every user of your web app at once.
Kulisa puts several browsers side by side in one window, each signed in as a different person. Be the buyer and the seller, the sender and the receiver, the admin and the user — and let an AI agent work through them while you watch.
Coming soon · free
Sam· seller
shop.example/orders
Order #12NEW
Order #11
Order #10
Elon· buyerclick “Checkout”
shop.example/cart
Checkout
Kim· admin
shop.example/admin
$ claude
> Elon pays for order #12, then I check that Sam sees it…
The problem
One app, many people, too many windows
When people use an app together, checking it means being all of them. A message is sent by one person and must arrive for another; an order placed by a buyer must show up for the seller.
Windows everywhere
A browser window per person, private tabs, a second browser for the third.
Signing in, again
Sessions expire, tabs get mixed up, and you sign in as the wrong person.
It all lives in your head
Which window is who, what you just did, and what should have happened.
How it works
Set the stage once. Then watch.
- The cast
A profile for each person
Create a profile for each person and sign in once, by hand. Each one keeps its own sign-ins, cookies and tabs, like a separate browser.
- The stage
Side by side, in one window
Their browsers sit next to each other. Send a message as one person and watch it arrive for the other.
- The wings
An agent you can watch
An AI agent runs in a terminal inside the window and can click and type in any profile. You see every step, and you can point at anything on a page to tell it what’s wrong.
Features
What Kulisa does today
A grid you arrange
Drag panes side by side, stack them as tabs, resize them. Pick a ready-made layout — profiles in columns, two by two, or one at a time — and Kulisa remembers yours.
See what the agent does
Every click and keystroke the agent makes is marked on the page, and a caption on the pane says what it is doing right now. Nothing happens off stage.
Point and tell
Click Pick, then click the thing on the page. The agent gets an exact reference to it, with a snapshot of that moment; you add what’s wrong. Pick in several panes for one message: “sent here, didn’t arrive there”.
The agent points back
When the agent wants to show you something, it outlines it right on the page, with a label on the pane, until you click or type there.
Works with the agent you already use
Claude Code is set up out of the box. Other agents that run in a terminal, such as Codex, can connect too. Kulisa doesn’t replace your agent; it gives it a stage.
macOS, Windows and Linux
One app, the same on all three.
Coming later
Next on stage
Timeline
Scroll back through everything every profile did, and point at the moment it went wrong.
A mailbox per profile
Each profile gets its own inbox for sign-up and invitation emails.
Save a run as a test
Turn what the agent just did across profiles into a test you can run again.
Profile reset
Start a profile fresh, keeping its name: handy for trying sign-up again.
FAQ
Questions
Is it free?
Yes. Kulisa is free.
When can I download it?
Soon. Installers for macOS, Windows and Linux are on the way, and the download buttons on this page will work as soon as they are ready.
Do I need to be a developer?
No. Anyone who works in a web app as several people at once — testers, support, product managers, people with several accounts — can use profiles side by side. The agent is there when you want it.
Which agents work?
Claude Code works out of the box: Kulisa starts it with everything it needs. Other agents that run in a terminal, such as Codex, can connect too.
Does the agent ever sign in for me?
No. A person always signs in, by hand. While a profile is on a sign-in page, the agent steps away from that profile until you are done.
Where is my data?
On your computer. Profiles, sign-ins and the things you point at stay on your machine; Kulisa has no account and no cloud of its own. What your agent reads is sent to the AI service that agent uses, as it would be anywhere else.
Which platforms?
macOS, Windows and Linux.
For developers: how does it work?
- Each profile is a persistent Chromium session of its own (cookies, storage, tabs), in an Electron window.
- The agent drives profiles through Kulisa’s MCP server: Playwright-based tools that take a profile id.
@playwright/mcpalso works through a per-profile CDP proxy, without a remote-debugging port. - Claude Code starts with a Kulisa plugin (MCP config, a skill on profiles, hooks). Other CLI agents run unchanged in a real terminal and find the server in
KULISA_MCP_URL. - A pick sends the agent a locator it can act on and a details file: outerHTML, key styles, a cropped screenshot, recent console errors and failed requests.
Everyone on one stage.
Coming soon · free