Capabilities

Useful browser work, inside clear boundaries.

Plan carefully, reuse what works, and verify browser effects without widening authority.

BrowserPair decision workbench representing careful planning and verified execution

Scope comes first.

Each agent key can be limited to specific browser identities, domains and capabilities. A task narrows that authority again before the extension acts.

01

Browser scope

Choose which connected browser identities an agent may use.

02

Domain scope

Restrict navigation and actions to the sites a workflow actually needs.

03

Capability scope

Grant only the browser action classes required for the task.

04

Approval policy

Pause for a person where the workflow or action risk requires it.

Brain plans. Reflex handles confident routine steps.

Brain chooses the next useful subgoal. Reflex can execute bounded routine actions when confidence is high; uncertainty routes back instead of guessing.

Same authority

Planning intelligence does not expand the browser permissions granted by the key or task contract.

Reusable knowledge without extra authority.

Skills capture where a workflow applies, its intent and repeatable stages. They help an agent reuse a known pattern while the same browser, domain and capability limits still apply.

Reusable stages

Keep repeated workflow structure available without teaching the whole process again.

Bounded execution

A Skill can guide work but cannot grant browser access the agent key does not already have.

Effects are checked, not assumed.

The extension verifies browser changes and returns durable task state. If BrowserPair cannot prove whether a side effect happened, recovery verifies or asks for resolution before another attempt.

Choose the capability, then choose the scope.

Connect a browser and give each agent only the access its workflow needs.

Open BrowserPair