Skip to content
Aghata

Browser mode

The work behind a login is still work.

Vendor portals, admin consoles, government forms, back offices with no API — Browser mode operates the real website in a real browser, visibly. Your credentials stay in a vault the agent can use but never read, and every use is written to a ledger.

A View through Three Arches of the Third Storey of the Colosseum, an 1815 painting by C.W. Eckersberg, pixel-dithered

The vault boundary

It can use your password. It can never read it.

How a session runs

Granted once. Worked visibly. Handed over when a human is required.

A browser session is not a script. It observes the page, decides, and acts — under the same plan, budgets, and approvals as every other Aghata session.

A View through Three Arches of the Colosseum (detail), pixel-dithered

Grant the credential once

Save a login to the vault and scope it — this time, this task, or this site. It's encrypted, substituted outside the model, and never appears in prompts or logs.

It operates the site, on the record

Real clicks in a real browser, with a live view you can watch and take over. Each step keeps evidence — except credential screens, which are never photographed.

It hands over when a human is required

A login wall or 2FA prompt pauses the run and makes the live view interactive. You sign in; the session resumes its plan. Your logins persist for next time.

Browser · live view

Trust, mechanically

Not promises. Mechanisms.

Every claim on this page corresponds to a structure in the product — a ledger row, a scope, a pause state — not a policy document.

Ledger

Every use is written down

Each time a credential is used, a ledger entry records which run, which site, and when. You review usage after the fact instead of guessing.

Scopes

Grants, not blank checks

Approve a credential for this time, this task, or this site. High-risk origins — banking, payments, SSO — always step up and re-ask, regardless of scope.

Evidence

Redaction by construction

Steps keep screenshots of what happened — but credential-bearing screens are skipped entirely. Auth pages are never photographed, so they can never leak.

Live view

Watch, or take the wheel

The session streams the actual browser. You can watch every click, and take over at any moment — it's your browser session, driven for you.

Persistence

Logged in stays logged in

A persistent browser profile keeps your sessions between tasks, so routine work doesn't restart from a login page every morning.

Rejection semantics

A rejected click cancels the run

Rejecting a risky browser action cancels the session rather than improvising around your no — because a refused step usually invalidates the plan.

The category

Where a chatbot stops and a script breaks.

Three ways to get software to work a website for you — and what each one actually does at the moments that matter.

Chatbot with browsingRPA scriptsBrowser mode
Sites without an APIReads public pages; can't act behind a loginWorks until the selector changesOperates the real site, logged in, adaptively
Who holds the passwordYou paste it into the chat — now it's in a promptStored in the script or a config fileA vault the agent can use but never read
At a login wall or 2FAGives upFails the jobPauses, hands you the keyboard, resumes
When something risky comes upNo concept of riskExecutes whatever was scriptedStops at an approval you can inspect
What's on the recordA transcriptA log file, if you built onePer-step evidence + a credential-use ledger

Questions

Questions about letting it log in.

Can the agent see my password?+

No — structurally, not by policy. Credentials are encrypted in the vault and substituted at the point of use, outside the model. Your password never enters a prompt, a step record, or a log. What the agent gets is a logged-in page; what you get is a ledger entry saying the credential was used.

What happens at a login wall or a 2FA prompt?+

The session pauses for authentication and the live view flips from watch-only to interactive. You log in — password manager, hardware key, whatever you use — and the run picks its plan back up. Aghata never tries to social-engineer its way past a second factor.

Can I watch it work?+

Yes — the live view shows the actual browser, click by click, and you can take over at any time. After the fact, every step keeps evidence: what it saw, what it did, and screenshots — except on credential screens, which are never photographed.

Does it log in fresh every time?+

No. Sessions keep a persistent browser profile per user, so a login granted on Monday still holds on Thursday's task. Grants are scoped — this time, this task, or this site — and high-risk sites like banking or payments always re-ask.

What if the site changed and it gets stuck?+

It stops and tells you where. Browser sessions run under an action budget; a run that can't make progress parks with its evidence trail intact instead of clicking in circles. Rejecting a risky browser step cancels the run — a rejected click usually invalidates the plan, and Aghata treats it that way.

Chapters of one system

Give it the outcome. Keep the control.

Start free in the product, or talk to us about how your team would use it.