AI BriefingBrowser capability added to eve agent
Key point
The agent-browser extension now lets the eve agent navigate and manipulate the web.
Details
The @agent-browser/eve extension allows the eve agent to navigate pages in the browser, read content, click, fill forms, take screenshots, and even inspect console and network activity. All operations run inside the agent's sandbox.
After installing with pnpm add @agent-browser/eve, mount the extension under agent/extensions/. The agent is provided with namespaced tools such as browser__navigate, browser__snapshot, browser__click, browser__fill, and browser__screenshot, and [ref=e12] references from snapshots are converted into @e12 selectors so that verified elements can be manipulated precisely.
Key features include the following.
- Sandbox isolation: Chromium and agent-browser run inside the eve sandbox, not the application.
- Domain allowlist:
allowedDomainscan restrict which domains are accessible for navigation and subresources. - Output and session control: Output size, screenshot handling, proxy usage, and session name can be configured via extension options.
- Credential protection: Commands related to cookies, storage, and saved auth state are not exposed to the model, and tool disabling or approval requirements can also be configured.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Google Releases Declarative Agent Orchestrator 'AX'