WebBrain field guide

Your browser agent has a lot of controls. You don’t need all of them.

Start with a page and a question. This guide shows the shortest path to useful results, then explains the deeper controls only when they become relevant.

Watch first · 12 min 40 sec

Settings, from the useful defaults to the advanced controls.

Start with the few choices most people need, then follow every Settings tab through providers, multimodal models, memory, skills, permissions, and the advanced drawer.

WebBrain 22.3.1 · 1080p · humane narration
First run

Your first 10 minutes

  1. Open a page you want help with. Click the WebBrain icon so its side panel opens next to that page.
  2. Stay in Ask mode first. Try “Summarize this page,” “What are the important dates?” or “Compare the plans in this table.” Ask can read, but it does not click or type.
  3. Switch to Act for a visible task. Try “Fill the search box with wireless headphones and run the search.” WebBrain will ask before consequential actions on a new site.
  4. Watch the run. You can stop at any time with the Stop button or Escape. Treat unexpected navigation, typing, or repeated actions as a reason to stop.
  5. Only visit Settings when you have a reason. The built-in WebBrain Cloud provider and recommended safety defaults are enough for a first run.
A useful prompt has an outcome

“Find the refund policy and tell me whether sale items qualify” is easier to verify than “look around this site.” Name the result you want and any boundary WebBrain must respect.

Pick the mode before the prompt

AskRead, explain, compare, extract. No page changes.
ActClick, type, navigate, download, schedule, and complete normal browser tasks.
DevAdd source, style, DOM, frame, and page-debugging tools. For technical work.

Ask is the calm default. Act is for work you could perform with the mouse and keyboard. Dev is not a “more powerful answer” mode; it is a specialized toolbox for inspecting and editing web pages.

A map, not a checklist

The six Settings tabs

You do not need to configure every tab. The color at the bottom of each block shows how often a typical person needs it.

Generalbehavior & comfort
Providersthe active model
Multimodalvision & speech
Memorywhat persists
Skillstrusted add-ons
Permissionswhere it may act
Everyday controlWhen you need itAdvanced trust decision
The safe baseline

Keep Plan before Act on Try, plan review on the confidence threshold, scheduled confirmations on, and Ask before consequential actions on. You can get very far without changing anything else.

Looking for internals?

The repository’s docs/ folder remains the source for architecture, providers, site adapters, prompt-injection defenses, localization, tool development, and test scenarios. The handbook you are reading focuses on what a person sees and controls.

Contributor docs

Architecture & tools

Architecture, adding a tool, and site adapters.

Security docs

Threats & data flow

Security model, privacy and data flow, and prompt-injection defense.