Ui.Vision has always combined classic web automation with local, fast and free computer vision and OCR. On top of that foundation sit large language models — and Version 10 turns that AI from a set of commands into an assistant: the AI tab in the side panel now builds and fixes complete macros for you. Thanks to the new free Ui.Vision AI tier it works right after installation, with no API key and no signup.


Ui.Vision AI in action

Turn an instruction into a reusable workflow

Create a macro: ask "Extract this table and save the rows as CSV." The assistant can inspect the page and build a macro using the uiv.* JavaScript API. Review and run it, then reuse it when the source data changes.

Repair a workflow: when a form or page changes, use Fix with AI to inspect the failure and help update the macro. Check the resulting behavior before using it for repeated work.

Keep classic automation: recorded actions, CSV-driven form filling and repeatable downloads remain useful. AI helps build the macro; you can still inspect and edit the automation in Ui.Vision.

Use an external assistant: connect through the MCP bridge to create, run and debug macros from your chosen MCP client. The bridge requires a local installation and connection to the extension. For native desktop steps or local file access, add the Desktop App / XModules.

The AI macro assistant: it writes and fixes your macros

Open the AI Chat tab in the Ui.Vision side panel and describe what you want in plain language — "create a macro that fills out this form" — or press Fix with AI on a macro that just failed. The assistant then works like a colleague sitting at your browser:

  • It inspects the live page (the DOM) to find locators that still work on the next run.
  • It writes the macro straight into the Macro tab, so you watch every command appear.
  • It runs the macro, reads the log and iterates on failures until the macro works — with a cap on fix loops, so it never spins forever.
  • It can capture element images itself and build image-based targets for uiv.findImage and visualAssert. The crop preview is shown right in the chat.
  • It explains each action in plain language, so you can follow — and learn from — what it does.

Your own macros are safe. Everything the AI creates lands in its own AI_Chat folder. The assistant never overwrites the macros you wrote yourself.

Curious how it knows every Ui.Vision command and all the real-world automation tricks? We publish the assistant's complete, unedited instructions: The Ui.Vision AI System Prompt. That page doubles as a compact best-practices handbook for web automation — cookie-consent banners, shadow DOM, OCR text targeting, error recovery — and it shows how to replace the prompt with your own in Settings > AI > System Prompt.

Free Ui.Vision AI — no API key, no signup

One-time setup: choose your AI. Use the free Ui.Vision AI service, or bring your own API key.
The one-time setup dialog in the AI Chat tab: pick the free Ui.Vision AI, or use your own AI.

The first time you open the AI tab, Ui.Vision asks you once which AI you want to use. Both answers are good answers:

  • "Use Ui.Vision AI — free, no setup needed" — we run the model for you. It is available on the free plan, and there is no signup and no API key to manage. Because it is a hosted service, your chat content (including the screenshots the assistant takes) is sent to our server for one purpose only: computing the AI answer. It is not stored. Hitting the limits of the free plan? Ui.Vision AI PRO gives you 10x the AI processing limits.
  • "I use my own AI — open settings" — bring your own API key, or point Ui.Vision at a model running on your own machine to keep AI processing local. See the providers below.

Your macros are stored and executed locally. AI generation and AI commands use your selected provider. Hosted AI sends chat content and any included page data or screenshots to that service for processing. A local model keeps that AI processing on your machine; websites, cloud OCR and other network services used by a macro still make their normal network requests. You can change your AI provider at any time in Settings > AI.

Bring your own AI: Anthropic, OpenRouter or a local model

Prefer to use your own account or your own hardware? Ui.Vision V10 supports three kinds of providers, both for the AI chat and for the aiComputerUse command:

  • Anthropic — Claude models. Create your API key in the Anthropic Console.
  • OpenRouter — one key, many models, including open-weight ones such as Qwen3 VL. Create your API key at openrouter.ai.
  • Local, OpenAI-compatible endpoints — run an open-weight model (again, Qwen3 VL is a good vision-capable choice) on your own machine and point Ui.Vision at it to keep AI processing local.

How to enter your own AI API key into the Ui.Vision RPA software
The AI settings tab

The AI commands for your macros

Next to the chat assistant, Ui.Vision has AI commands that you can use inside a macro, so AI becomes one step of a larger automation:

  • aiPrompt — send any text prompt to the LLM and store the answer in a variable. Added in V9.3.6.
  • aiScreenXY — describe an element in words and get its X/Y screen coordinates back. Added in V9.3.6.
  • aiComputerUse — hand a whole task to the AI agent and let it drive browser and desktop. Added in V9.3.8.

AI chat, Claude and Computer Use demo macros in Ui.Vision RPA
Test Computer Use and the other AI commands with the preinstalled demo macros.

More documentation — for humans and for AI

AI support in Ui.Vision V10 keeps evolving. Based on your feedback in the user forum, we will improve and expand the integration with the next updates.

Anything wrong or missing on this page? Suggestions?

...then please contact us.

Fresh from the Ui.Vision Forum: The Latest 3 Topics.

← Meet the Ui.Vision team and users in our forums.