How it works

One control layer between an agent and your browser.

BrowserPair lets an AI agent request bounded browser work without exporting your signed-in browser profile or giving the agent unrestricted control.

The flow

An agent never talks directly to page internals. BrowserPair inserts a durable, scoped control layer between the agent and the browser profile you already use.

Connect a browser profile

The extension pairs the current browser with your BrowserPair account. Existing sessions stay in the browser.

Create an agent key

Give each agent a separate key scoped to browser IDs, domains, capabilities and an approval preset.

Send a bounded task

The agent uses Remote MCP or the REST API. Every task carries an immutable contract describing what domains and capabilities are allowed.

Execute inside the extension

The extension observes the selected page, performs allowed actions, verifies effects and returns durable task state.

Permissions come from you, not the page.

Page content is treated as untrusted data. It cannot expand the authority granted by your account, agent key or task contract.

01

Scoped agent keys

Restrict an agent to specific browsers, domains and capabilities. Revoke or rotate one agent without disturbing another.

02

Approval presets

Choose when the product should pause for confirmation, from always ask through bounded low-risk automation.

03

Hard capability boundaries

Approval presets never add capabilities. The hard scope remains the ceiling for the key and task.

04

Sensitive actions stay explicit

Publish, send, delete, payment, password and transfer effects require the right capability and policy-required confirmation.

The browser stays in control of mutations.

BrowserPair Cloud coordinates identity, permissions, routing, usage and audit. Browser mutations happen only inside the extension executor on the connected device.

Important boundary

The cloud does not become a remote browser. It routes a scoped task to the extension that is connected to your browser profile.

Strong tab ownership

Hosted tasks act only on task-owned tabs. BrowserPair does not silently adopt an arbitrary existing tab.

Redacted sensitive values

Passwords, OTPs, card data, API keys, auth tokens and cookies are redacted before observations leave the device.

Bounded primitives

Agent actions are constrained to browser operations such as read, click, fill, select, navigate, press and wait rather than arbitrary page JavaScript.

Retries do not mean blind replay.

Task state is durable across reconnects. BrowserPair uses idempotency, ownership generations and effect verification to avoid duplicating side effects.

Effect unknown

If BrowserPair cannot prove whether a side effect happened, it verifies or asks for resolution before allowing a new attempt. It does not blindly run the effect again.

That same recovery model lets a browser reconnect and continue the same task identity instead of inventing a second task.

Ready to pair an agent?

Connect your browser, create a scoped key and choose REST API or Remote MCP.

Open BrowserPair