Use a text grid to let agents handle web forms

They made a browser view that turns pages into numbered text so an agent can work through applications without image models.

Done with Custom agent· Advanced· Feb 2026

What they did
A developer built TextWeb, which opens a site in headless Chromium and records each visible item’s location, text, and whether it can be used. It then redraws the page as a fixed character grid that preserves the layout, with actionable items assigned numbers. An LLM can read that text and operate the page with simple numbered instructions.
How it went
The text version of a page is only about 2–5 KB, while a screenshot is around 1 MB, so the setup avoids the need for a vision model. The project did not report finished job applications and instead asked for reactions to the grid design.
Tips
This can be useful when you want an agent to work through forms using plain text rather than screenshots. It can be connected with an MCP server, function-calling definitions, LangChain, CrewAI, an HTTP API, a CLI, or a Node.js library.

Try it yourself with Claude Code

Act as a text-based browser for [site or form]. Show me the page as a numbered text grid, keeping the visible layout as closely as you can. Let me inspect the grid first, then help me enter [details to fill in]. If the next step would submit, pay, delete, send, or otherwise finalize anything, stop and ask for my approval before continuing.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.