Ankita v2.4.3: a live browser beside your chat

Ankita, an open-source desktop AI assistant, has released version 2.4.3, introducing a live browser feature that allows the assistant to interact with web pages directly alongside chat conversations. This new capability enables the assistant to open web pages, read snapshots of their contents, click, type, scroll, and fill forms.

The browser stage is built into the desktop app, with tabs, URL, page status, and current page visible, along with Stop and Take control buttons. The assistant can also fill up to ten form fields from a single page snapshot, including selections and checkboxes, while protecting passwords from being exposed in the conversation.

There are two browser plugins, both of which start disabled and can be opted-in by the user. Playwright Browser runs in its own separate Chromium profile, ensuring that the assistant does not interfere with the user's personal browser session. Chrome local connects to Chrome through a pinned Chrome DevTools MCP server, offering options for a private session, a debug port, or the user's own session.

Setup includes progress tracking for the Chromium download, background mode, and a debug-port connection test. Every plugin has allowed and blocked domain lists, and navigation is restricted to HTTP/HTTPS only, with private hosts blocked by default.

Browser actions such as clicks and form submissions require approval from the user, while reading pages does not. The assistant will not automatically replay actions if a connection hiccup occurs, instead requesting confirmation to prevent potential issues. Performance and filesystem security enhancements include running command creation, shell discovery, and process-tree work in a worker process on Windows, coalescing terminal Markdown redraws on a fixed 20 ms deadline, and measuring file inspection costs incrementally rather than re-scanning the entire conversation.

To enhance security, file tools enforce workspace boundaries for paths, reject traversal escapes, interior symlinks/junctions, Windows device aliases, and alternate data streams, and use exclusive, unpredictable staging files for atomic writes.