Web task router / 06

The web keeps moving.
Your agent needs more than one way through it.

Read with clean extraction, navigate with reusable site knowledge, and open a real browser only when the task requires interaction. The agent chooses a surface that matches the job.

https://supplier.example/portal/orders/1842
Live browserClean extractionSite guidanceTask #1842 · running
SUPPLIER PORTALOrdersInventoryAccount
ORDER #1842

Confirm revised delivery

Dock availability changed. Select a new appointment window.

Aug 14 · 09:00Aug 14 · 14:30 ✓Aug 15 · 11:00
10:14:02 page ready10:14:04 option selected10:14:04 approval requested
Choose by page behavior

Three surfaces.
One task contract.

Browser agents often use one expensive interaction loop for reading, navigation, and submission. That increases latency and exposes the agent to more untrusted page content than necessary. Pilot’s complementary web apps let a workflow begin with clean extraction, move to typed guidance, and open an interactive browser only for the steps that require it.

Route 01Read

Web pages as clean Markdown

Plainweb turns a URL into compact content for extraction, summarization, and downstream research.

Dynamic
Route 02Act

A real Chrome execution surface

Otto connects the agent to browser tabs for extraction, screenshots, and supported site interactions.

Readable
Route 03Repeat

Task-specific site guidance

Bowmark stores concise navigation knowledge and improves it with feedback from completed workflows.

Repeatable
Web workflow router

Route one brittle workflow
through the right web surface.

Give us the site and task. We will separate what should be extracted, what can follow known navigation, and what genuinely requires a live browser.

What you get
  1. 01Plainweb/Bowmark/Otto split
  2. 02Interaction boundary
  3. 03Proposed result artifact
Start with one real workflowAbout 2 minutes

Do not include credentials, secrets, or customer data. By submitting, you agree that Pilot may use this information to respond. Privacy Policy.

Task trace

From intent to observable result.

A content question should not launch a browser. A known workflow should not rediscover the site. A genuine interaction can escalate to Chrome with the relevant context already assembled.

  1. 01

    Read before opening a browser

    Use clean extraction for content-only tasks and reduce unnecessary interaction.

  2. 02

    Load task guidance

    Apply a known site path when the destination and workflow are understood.

  3. 03

    Escalate to Chrome

    Use the interactive surface for forms, authenticated sessions, or visual state.

  4. 04

    Capture what worked

    Feed successful task knowledge back into a reusable navigation guide.

Technical basis

Specific product surfaces.
Inspectable in the docs.

Product reviewPilot release 1.13.9Inspect documentation
  • 01Clean URL-to-Markdown extraction
  • 02Real Chrome extension and relay
  • 03Screenshots and page extraction
  • 04Reusable site-task guidance with feedback