Run browser work with your AI agent
Nextbrowser is a desktop workspace for local AI agents and managed browser sessions. Choose a browser identity, give your agent a task, and keep the work visible from setup to result.
Everything around the agent run
Nextbrowser brings identity, browser state, agent controls, and review into one operational view.
Profiles and sessions
Keep browser identities and their sessions separate for different accounts, customers, regions, or test runs.
Proxy and fingerprint rotation
Refresh a browser identity when a workflow depends on geography or a clean, isolated session. Verify the result before continuing.
Agent workspace
Run tasks with chat history, queues, stop and edit controls, streamed activity, and conversation forks in one desktop workspace.
Skills and custom scripts
Reuse browser workflows and let preflight prepare the selected profile, session, tab, and target page before a run.
Scheduled runs
Set up recurring tasks with an agent, prompt, time, weekdays, and an enabled state. Scheduled runs work while the app is open.
Live View and diagnostics
Watch the active page and inspect status, tabs, screenshots, page state, and identity checks while an agent works.
Connect the agent you already use
Nextbrowser provides the browser context and the visible work surface. The selected agent still needs to be installed and authenticated on your computer.
Claude Code
Nextbrowser connects to Claude Code through its CLI installation.
Installation guideCodex
Nextbrowser connects to the Codex executable bundled with the ChatGPT desktop app.
Download ChatGPT for desktopYour first browser run
Set up a focused browser context, then keep the result observable while the task runs.
- Install Nextbrowser. Download the latest macOS or Windows build from GitHub.
- Sign in and pair the app. Continue in your browser and approve the pairing request.
- Create or select a profile. Use a separate profile when accounts or browser state must stay isolated.
- Start the session. Open the target page and confirm the active tab in the sidebar.
- Connect your agent. Select an installed Claude Code or Codex runtime.
- Send a narrow task. State the goal, limits, and expected result. Watch Chat and Live View as it runs.
Turn a task into a workflow
Use reusable instructions for repeatable browser work, with a preflight check before dispatch.
Review the skill or custom script and remove secrets or stale assumptions.
Choose the profile, target site, and local agent you intend to use.
Let preflight confirm the session, inspect tabs, activate or open the target page, and wait for readiness.
Supervise the run, stop or edit it when the task changes, and verify the final page before any consequential action.
Purchases, publishing, account changes, deletions, and other consequential actions should stay under your control.
Check the browser context first
Most run failures become easier to locate when you check dependencies from the outside in.
Stop at the first failed check. A missing tab or stopped session can look like an agent failure later in the run. When reporting a problem, include the app version, operating system, expected behavior, actual behavior, minimal steps, and redacted logs or screenshots.
AGPL-3.0 for Nextbrowser
Nextbrowser is available under the GNU Affero General Public License v3.0 only. Commercial use, modification, and redistribution are allowed.
If you distribute a modified version, or make a modified version available for remote network interaction, you must offer the corresponding source under AGPL-3.0-only. This applies to Nextbrowser itself. Dependencies and connected tools remain under their own licenses.
Read the full license