Task-owned workspaces
Agent work is organized into visible task workspaces in your existing signed-in Chrome profile.
Local browser runtime for AI agents
Any agent. Your browser. Your rules.
Give an agent a tab, not the keys to your browser.
Unreleased: AgentTab v2.0.0-rc.1 is under review. No v2 installation is available from this site.
Needs your attention
Complete a password, passkey, or other human-only input.
The agent keeps working and verifies the result itself.Built for useful work without silent reach
Agent work is organized into visible task workspaces in your existing signed-in Chrome profile.
Attention notices hand passwords, passkeys, 2FA, CAPTCHAs, payment secrets, and similar input back to you.
Commit presents a review step for recognizable sends, publishing, purchases, deletion, uploads, authorization, and permission grants.
Outcome-first browser use
AgentTab makes the ordinary path clear: a task gets a workspace, the agent works in it, and people take over when an action needs them.
A task opens or adopts an explicit tab. Ownership is visible and stays scoped to that task.
Standard actions use the task workspace, not broad browser control.
Attention notices protect secret and other human-only input without pausing the task. Recognizable consequential actions wait for Commit.
Trust boundary
Task ownership is an execution and coordination boundary. It is not cookie, identity, or profile isolation.
Standard mode does not provide raw cookies, storage, passwords, arbitrary JavaScript, raw Chrome DevTools Protocol access, or coordinate actions.
Private by architecture
AgentTab does not collect, sell, or share personal data. Browser work happens through your local runtime and your own signed-in browser.
Read the full privacy policy