How It Works (Simple)

Unbrowse turns one real browsing session into reusable automation.

It captures a browser-authenticated flow once.

Then other agents can reuse it by intent.

Visual Overview

Step 1: Capture (HAR)

You (or an agent) uses the site normally.

We capture a HAR trace of network traffic.

We redact secrets.

We keep request/response structure and sequencing.

Step 2: Learn + verify (publish a skill)

We extract candidate endpoints from the HAR.

We normalize URL templates and parameters.

We validate request/response schemas.

We publish a versioned skill with verification proofs.

Step 3: Discover by intent

When an agent asks for something:

1. We resolve intent against the discovery index.

2. We rank candidates by reliability, freshness, and domain match.

3. We select the best skill version for execution.

Step 4: Execute locally with browser parity

Execution happens in a browser context when needed.

That preserves:

• cookies

• CSRF state

• OAuth redirects

• strict browser-only behavior

Step 5: Safety + feedback

Reads can run automatically when confidence is high.

Mutations require explicit confirmation.

Every run produces a trace id.

Feedback updates reliability and can trigger re-verification.

Billing and payouts are disabled in the current beta.

Getting Started

Indexers

Start with Getting Started.

Developers

Start with Quick Start.

---

Next: Explore Key Concepts to understand the terminology and technical foundations.