CodeNeterBrowser
Preview and debug web apps in the CodeNeter desktop browser, add page context to a conversation, and let the agent navigate, interact with pages, and read console errors.
The CodeNeter desktop app includes a browser for web development. Use it to preview a local dev server beside the conversation, inspect parts of the UI, and give the agent visual and page context. The agent works in the same browser: it loads pages, reads and interacts with them, takes screenshots, and checks the console, so it can look at the result of a change it made or reproduce a problem you describe. Browser tabs open from the workspace tab bar, and each conversation has its own.
Browser capabilities
Sign-ins, permission prompts and other browser dialogs, file uploads and downloads, and content inside frames require user interaction. For those, the agent asks you to take over.
Element context
Inspect element lets you point at part of the page and describe the change you want. The selected element is attached to your message with page context to help the agent find the relevant code. You can include several elements in one request, for example to ask for consistent spacing across a group of buttons.
With an image-capable model, the agent also receives a screenshot highlighting the selected element. Add screenshot to chat shares the visible browser area instead. Design with the agent covers the full workflow.
The page's developer tools are also available while you work.
Responsive preview
Responsive mode shows the page at other viewport sizes, with rotation and a fit-to-window view. When checking layouts, the agent can resize the viewport and emulate light or dark color schemes and reduced motion. It restores those settings when it's done.
Common uses
- Check a UI change after the agent edits the code.
- Reproduce a reported bug and read the console error behind it.
- Review a layout at narrow widths or in dark mode.
- Point at an element you want restyled or fixed.
Control and approvals
Browser actions are MCP tool calls, so they follow your MCP approval setting and ask before running by default.
While the agent works in a tab, the page shows that the agent is in control. Taking over, or clicking, typing, or scrolling in the page, stops browser automation for the rest of that run. The conversation continues, and the agent can use the browser again on your next request.
Stored data and isolated tabs
Regular tabs share one persistent store of cookies and site data, which remains after you quit the app. Sites you sign in to stay signed in when the agent uses the browser, so its actions there act on your account.
For work that shouldn't touch stored data, such as testing against mocked APIs, the agent can open an isolated tab. Its cookies and storage stay in memory and aren't shared with other tabs.
Context and compaction
Point the agent at the right files, images, and page elements, track context use, and keep long conversations going.
Terminal
How the CodeNeter agent runs shell commands in your project, handles long-running and background tasks, and how the desktop app's integrated terminal differs.