Key Concepts

Understanding these core concepts will help you grasp how Unbrowse works.

Skills (Abilities)

Definition: Reusable, executable actions learned from real browsing.

In the platform, these are stored as versioned skills.

Agents often refer to them as abilities.

Think of abilities like recipes:

• A recipe tells you how to make chocolate chip cookies

• An ability tells an AI agent how to "post to LinkedIn" or "book a flight"

Key characteristics:

• Captured from real browsing: Derived from HAR traces

• Reusable: Other agents can reuse the same flow

• Request-level execution: No pixel-based clicking required

• Browser parity when needed: Can execute inside a browser context to preserve cookies and CSRF

• Versioned: Each skill has immutable versions and an audit trail

• Parameterized: Accepts inputs (dates, search terms, etc.)

Examples:

• "Search for flights on United Airlines"

• "Post update to LinkedIn with image and tags"

• "Check account balance on Chase Bank"

• "Create Jira ticket with priority and assignee"

• "Pull competitor traffic data from SpyFu"

How they're created:

1. User browses website normally with extension installed

2. Extension captures network traffic (HAR)

3. Candidate endpoints are extracted and normalized

4. Verification runs validate schema and replayability

5. Skill is published and indexed for discovery

Endpoints

Definition: The executable request templates inside a skill.

Each endpoint describes:

• Method + normalized URL template

• Headers/query/body templates

• CSRF/OAuth replay plan (when needed)

• Idempotency (safe vs unsafe)

• Verification status + reliability score

Discovery Index

Definition: The routing layer that matches intent to the best skill version.

It ranks by:

• Intent similarity

• Domain/subdomain match

• Reliability and freshness

• User feedback

• Safety and abuse risk

Indexers

Definition: Users who contribute browsing patterns to the network.

Who can be an indexer:

• Anyone with a browser

• No technical skills required

• Just browse normally

What indexers do:

1. Install Unbrowse browser extension

2. Browse websites as they normally would

3. Extension captures network patterns automatically

First-mover advantage:

• The first person to index "Book flight on United" captures that workflow

• Early indexers benefit most from high-value workflows

Examples of valuable indexing:

• SaaS tools (Notion, Slack, Jira, Salesforce)

• Financial platforms (banking, trading, crypto exchanges)

• Data sources (analytics, research, market data)

• Social media (LinkedIn, Twitter, Instagram)

• E-commerce (Amazon, Shopify, marketplaces)

Large Action Model (LAM)

Definition: The AI brain that analyzes captured browsing patterns and generates executable abilities.

What LAM does:

1. Pattern Analysis

   • Studies millions of captured browsing sessions

   • Identifies common interaction patterns

   • Understands authentication flows

   • Recognizes data structures

2. Ability Generation

   • Converts HAR files into executable API wrappers

   • Generates parameter schemas

   • Creates error handling logic

   • Optimizes for reliability

3. Workflow Composition

   • Chains multiple abilities together

   • Handles data flow between steps

   • Optimizes execution order

   • Manages dependencies

4. Routing & Matching

   • Matches natural language queries to abilities

   • Selects best ability variant for each request

   • Learns from success/failure patterns

   • Improves recommendations over time

Network-Level Execution

Definition: Direct HTTP communication with websites, bypassing visual UI entirely.

How it works:

When you click a button on a website:

[You click "Buy Now"]
      ↓
[Browser sends HTTP POST request to server]
      ↓
[Server processes order]
      ↓
[Server sends response back]
      ↓
[Browser displays confirmation page]

Traditional automation forces agents to:

1. See the button visually

2. Figure out where to click

3. Simulate the click

4. Wait for page to load

5. Parse visual result

Unbrowse lets agents:

1. Send the HTTP request directly

2. Receive the response

3. Done

No screenshots. No clicking. No waiting.

Why this is revolutionary:

• Speed: seconds (GUI) → sub-second (network)

• Reliability: brittle selectors → stable request patterns

• Maintenance: UI changes break automations → patterns tend to survive UI changes

Technical benefits:

• No browser overhead

• No rendering engine needed

• No visual parsing

• Direct server communication

• Structured data responses

• Parallelizable (run 1000s simultaneously)

Browser-parity execution

Some sites bind authentication to real browser behavior.

Examples include strict CSRF, same-site cookie policies, and OAuth redirects.

For those, Unbrowse can execute requests inside a browser context.

This keeps the speed of request-level execution.

It preserves login state and browser-specific behavior.

Ability Index

Definition: The searchable database of all available abilities.

Think of it as "Google for agent actions":

• Google indexes information (web pages)

• Ability Index indexes actions (how to do things)

How agents search:

Natural language queries:

• "How to book flight on United"

• "Post to LinkedIn with image"

• "Check balance on Chase"

---

Next: Learn the workflow in How It Works (Simple).