on-device
any website
24 × 7
→ Kiwi Scout

Real work,any browser,24×7.

Kiwi Scout is a virtual computer agent that completes web tasks end to end, intelligently. No APIs needed. On-device by default.

kiwi-scout / run · on-device● running
Point it at a task. It finishes the job and shows its work.No APIs · no plugins
Agentic workflowsEnd to end

Point Scout at a task. It finishes the job.

And it shows its work: every step, screenshot and diff, with a pause for your approval where it matters.

01

Works across any website

Automate tasks directly in the browser, no APIs or plugins needed. Scout handles forms, logins, cookies and pop-ups, and adapts when the UI changes.

FormsLoginsCookiesPop-upsUI drift
supplier portal · /login
Sign inops@company.com••••••••••Sign in
  • Cookie banner · rejected non-essential
  • Logged in · saved profile
  • Pop-up closed
  • Layout changed · re-anchored
ChromeEdgeBraveNo APIs · no plugins
02

Schedules and runs while you sleep

Daily, weekly or event-based triggers for tasks like stock price snapshots, flight fare scans and tariff lookups. Scout runs them on its own.

DailyWeeklyEvent triggers
Price snapshot · watchlist
Every weekday · 06:00 next: tomorrow
  • Mon 06:00 34 values
  • Fri 06:00 34 values
  • Thu 06:00 33 values · 1 retried
03

Review before it ships

Pause for approval at key steps, compare diffs and screenshots, then continue, edit or retry in one click.

ApprovalsDiffsScreenshotsRetry
  1. Export supplier list214 rows · 2 min
  2. !Review before it ships3 rows changed since last run − Acme Parts · net 30+ Acme Parts · net 45ContinueEditRetry
  3. 3Upload to supplier portalwaiting for approval
04

Store data your way

On-device by default. Export to CSV, Google Sheets, web uploads or API handoffs when you choose.

On-deviceCSVGoogle SheetsAPI
Local file
Web form
Cloud sheet
On-devicedefault storage
Export →CSVGoogle SheetsWeb uploadAPI
05

Many tasks, many browsers

Run tasks in parallel across browsers, spread steps across windows, isolate sessions by profile, and set per-browser rules and limits.

Parallel runsProfilesPer-browser rules
4 tasks · 4 browsersisolated sessions
Finance profile
Fare scan3/7 · running
Ops profile
Tariff lookup26/40 · running
Research profile
Price snapshots34/34 · done
Personal profile
Form fill1/5 · paused · max 2 tabs
02:00 · 03:00 · 04:00

It runs while you sleep.

Fare scans at 2 a.m. Price snapshots before the open. Tariff lookups every Monday. Scheduled, checkpointed, resumed after blips, waiting for your review in the morning.

Built inReliable at scale

Built to run for hours, not minutes.

The capabilities that keep agentic workflows reliable when the pages are messy, the files are large and the run is long.

01

Vision-grade extraction

Reads unlabeled charts, dashboards and canvases with multimodal vision. Pulls KPIs and values from images, PDFs and dynamic, non-DOM interfaces.

02

6–8 hour endurance

Built for long-running automations: download, transform, upload. Streams large files, checkpoints and resumes after blips, with backoff and retry in the background.

03

Model-agnostic

Swap frontier or local models for the best price and performance. Route by task, cost and latency, with fallbacks and on-device options.

04

One click to the Kiwi platform

Hand results to orchestration graphs or API integrations with no glue code. Trigger downstream jobs or send via webhooks.

Dashboard · canvas, no labelsread by vision
PeriodValueΔ
P458.0+12
P563.0+5
P671.0+8

Reads what the DOM can’t.

Charts with no labels, dashboards drawn on canvas, values trapped in images and PDFs. Scout reads them the way a person does, then pulls the numbers out.

ChartsCanvasesImagesPDFsNon-DOM UI
The filmScout at work

Remove grunt work. Forever.