How Local AI Could Change the Game
Introduction
AI tools are evolving quickly. Every week introduces new features, and it can be hard to know what’s truly worth your time. But occasionally, someone in the accounting tech space builds something genuinely useful. This time, it's Beau Gaudron from Growthwise.
Beau has been working on a new kind of browser-based automation using a large language model (LLM) that runs entirely on a local machine. The tool can interpret browser content and complete tasks without sending any data to the cloud. For firms that need to keep sensitive client data in-house, this is an important shift. It offers the benefits of AI-powered workflows while keeping privacy and control firmly in your hands.
What Beau Built
After reading about OpenAI’s Atlas and other browser automation tools, Beau began exploring how to build a local version. His goal was to connect a web browser to a local AI model that could understand screen content and carry out commands. The result is a working prototype. It reads the page, processes natural language instructions, and performs browser-based actions on its own. All of it runs locally on the user’s device.
This means accountants can now automate browser tasks without relying on third-party services or risking client data leaving their systems.
Smarter Than Scripts
Traditional robotic automation relies on fixed steps: click here, enter that, then submit. These workflows often break if a button moves or a page layout changes. AI handles this differently. It processes the browser in context, reads labels, identifies relevant actions, and adapts as it goes. This makes it far more robust in real-world use.
“Robotic browser automation is a great starting point,” Beau said. “But the next iteration is being able to task an AI to interpret and complete actions based on our browsing results.” Rather than following rigid scripts, the AI behaves more like a person navigating a page. It sees what’s there and makes decisions accordingly.
What It Looks Like in Action
In a demo, Beau showed two examples using Xero:
• In the first, he gave the AI a simple instruction file:
• Open the profit and loss report
• Open the export menu
• Click PDF
The AI took over the browser and executed each step in order. The task was completed successfully, with only minor pauses as the model processed what to do next.
In the second test, Beau gave a more general command: “Show me all contacts.” This time there were no predefined steps. The AI had to interpret the request and find the correct section of the platform. It paused slightly longer but still completed the task. Everything ran on Beau’s machine. No data was sent to external servers.
Making Web Pages AI-Friendly
To help the AI understand browser content more effectively, Beau used the browser’s accessibility tree view instead of raw HTML. The accessibility view presents a simplified version of the page, which screen readers often use. It’s cleaner and easier for an LLM to interpret.
“When we use the accessibility tree view, it's much more readable to a human, which means it's also much more readable to an LLM,” Beau said. This gives the AI a better way to match instructions to visible page elements, improving both accuracy and flexibility.
Why Local AI Is a Big Deal
Many firms hesitate to adopt AI tools because of privacy concerns. Cloud-based systems often involve sending client data to external platforms, which may not meet internal or regulatory standards. Beau’s local model avoids that completely.
“It’s not going out to OpenAI. It’s not going out to Grok. It’s just all running locally,” he said. This means firms can safely test and use AI for browser tasks without exposing sensitive data. It also opens up automation for firms with strict compliance obligations.
Practical Use Cases for Firms
Most accounting teams spend hours on repetitive tasks in web apps. These include:
• Exporting reports
• Downloading and renaming files
• Updating client records
• Uploading documents to portals
• Performing monthly checks
With local AI, these tasks can be delegated to a browser assistant that works in the background. It is not about removing people from the process. It is about reducing the time spent on low-impact admin so your team can focus on higher-value work.
Imagine typing, “Export all quarterly reports and save them to the client folder,” and watching the AI do it for you.
What About Cloud-Based Tools?
Beau was clear that not every firm needs to run local AI. Some will prefer cloud tools for their ease and setup speed. “If you just want a turnkey, robust solution and are happy with a cloud provider, the one that surprised me the most in my research was Anchor Browser,” he said.
Platforms like Atlas or Comet also offer fast deployment and scale. For firms that prioritize control or want to trial AI without long-term commitments, local tools may be the smarter entry point.
AI That Takes Action
This project is still early, but it reflects a larger trend. AI is no longer limited to analysis. It is now capable of action—clicking, navigating, and executing tasks within the same tools firms already use.
“It’s not foolproof at all,” Beau admitted, “but it’s a really good way to interact with the browser and not have to have a whole heap of compute resource.” That focus on progress over perfection is what makes it worth paying attention to.
Where to Begin
This technology is already usable. It does not require a full AI strategy or a big investment. It just needs curiosity and a willingness to experiment.
Start with one task. Choose something low risk and repeatable. Run it locally. Measure the time saved and refine as you go.
The future of automation in accounting is not only in analyzing data. It is in acting on it securely, contextually, and within your own systems. This is not a concept. It is real. The question now is which firms are ready to test it.
