You're offline, showing the last version we saved.

Part of our Xero coverage. More Xero news →

Last week, Beau Gaudron - Tech Ninja at Growthwise - did something most of us in accounting only talk about: he actually built a working AI tool to automate part of the job.

Not some high-budget, VC-backed platform. Not an enterprise app rollout. Just a weekend project, using open-source tools, a locally hosted AI model, and a Chrome extension.

The tool? It reconciles Xero transactions in your browser - using a local LLM (Large Language Model) that doesn’t rely on any paid subscription or cloud-based AI.

It’s smart, secure, and most importantly - it works.

Where it all started

The spark came from a post by Enzo O’Hara Garza shared online and resurfaced by Natalie Lennon FCA.

The original post floated the idea of reconciling Xero transactions with the help of a Chrome extension - something that caught Beau’s attention immediately.

It wasn’t just theory. The idea was also mentioned in a recent AutomationTown podcast episode with Chad Davis. That gave Beau the push to take it further.

So, like any curious tech-leaning accountant, Beau turned it into his weekend build.

No fluff. Just a working prototype that solves a very real accounting workflow problem.

What Beau’s Chrome extension actually does

Let’s break down what the tool does - and why it matters.

Beau’s extension connects with your live Xero file via the browser. Once connected, it scans through unreconciled transactions in Xero, feeding the data into a locally hosted AI model that runs entirely on your machine.

From there, the AI model:

• Reviews and checks each transaction

• Suggests potential matches

• Flags those transactions for approval

• Recommends whether to reconcile - or not

The entire process happens without any data ever leaving your device. That’s a huge win for firms who are wary about sharing sensitive financial data with cloud AI tools or third-party providers.

It’s a clean, lightweight experience - and surprisingly efficient, especially considering the build was completely DIY.

The tech stack behind the build

Beau’s setup is refreshingly accessible. He didn’t use any proprietary software or paid AI platforms to power the logic. Here’s what made it work:

• Claude Sonnet 4 – for help refining the code logic

• OpenAI GPT-4o – to guide him step-by-step through the Chrome extension build

• VS Code – where he developed and tested everything

• Chrome – where the extension lives and runs

• Xero – the accounting software at the centre of it all

• Ollama AI + Mistral 7B – the local model powering the reconciliation logic

That last one is especially important. Mistral 7B is a relatively lightweight open-source model that can be run locally on consumer-grade machines. And Ollama makes it possible to use it in a command-line environment without needing a massive GPU setup.

This combination gives you something incredibly powerful: AI-powered accounting automation without the need for internet access, monthly fees, or cloud privacy concerns.

Why this matters for the accounting industry

Beau’s tool is more than a cool side project - it’s a signal. A signal that accounting professionals aren’t just passive recipients of tech. We can build it ourselves.

Here’s why this matters:

1. Local AI is real - and practical

For a long time, local LLMs felt like a science project. Too slow, too unstable, too hard to install. This project proves otherwise. With Mistral and Ollama, Beau’s showing that you can run AI workloads privately and quickly without needing high-end hardware.

2. AI doesn’t need to be expensive

There’s a belief that using AI in accounting requires ChatGPT Plus subscriptions, enterprise contracts, or integrated SaaS platforms. Beau just proved that’s not true. His solution costs $0 to run once set up.

3. Accountants can (and should) build the tools they need

Beau’s not a full-time dev working in Silicon Valley. He’s a tech-leaning accountant inside a firm. That context matters. He understands the problems, the workflows, and the risks. That’s why this works - because it’s solving a real problem from inside the profession, not outside it.

4. This is where innovation starts

Innovation doesn’t start with a product launch - it starts with curiosity. With someone saying “what if I could make this easier?” And Beau’s shown just how far you can go when you follow that question with action.

What it means for other firms

If you’re in public practice and thinking about AI, Beau’s project gives you a blueprint to follow:

• Start small

• Solve something real

• Use free and open tools

• Learn in public

• Share what you build

There’s no need to “wait for the tech to mature” or hope a software vendor builds the perfect feature for your firm. The tools are already here - and they’re accessible to anyone willing to experiment.

This also opens the door to custom, local AI workflows inside your firm - ones that never touch the cloud and don’t need to be bundled into a platform.

Want to try it or see how it works?

Beau was generous enough to share the full build walkthrough on YouTube, where he goes through everything:

• How the Chrome extension connects to Xero

• How the AI model is hosted and run locally

• The full logic and flow of how transactions get matched

• How you can test it and customise it

Watch on Linkedin ➔ Beau’s original post

A note of thanks - and a call to action

Massive credit to Beau for leading the charge here. This isn’t a fluff LinkedIn post or an AI-generated “what if” - it’s a working tool, built by someone in the industry, for the industry.

If you’re an accountant or bookkeeper who’s been sitting on an idea, watching others experiment - let this be your sign. The barrier to entry is lower than ever. And the opportunity to build something useful has never been more accessible.

Go follow Beau, watch the video, and start experimenting.

Because this - DIY, practical, privacy-first automation - is where the next era of accounting tech is going to come from.

Frequently asked questions

What does Beau’s Chrome extension do?

It connects to a live Xero file through the browser, scans unreconciled transactions and passes them to an AI model running on the user’s own machine. The model reviews each transaction, suggests potential matches, flags them for approval and recommends whether to reconcile.

Does client data leave the computer?

No. The reconciliation logic runs on a locally hosted model, Mistral 7B served through Ollama, so transaction data never leaves the device. That is the main draw for firms wary of sending sensitive financial data to cloud AI tools or third-party providers.

What tools did Beau use to build it?

VS Code for development and testing, Chrome to host the extension, and Xero as the ledger. He used Claude Sonnet 4 to help refine the code logic and OpenAI GPT-4o to guide him through the Chrome extension build, with Ollama and Mistral 7B powering the reconciliation itself.

Does it need powerful hardware or a paid AI subscription?

No. Mistral 7B is a relatively lightweight open-source model that runs on consumer-grade machines, and Ollama runs it from the command line without a large GPU setup. Once set up, the tool needs no internet access and no paid AI subscription.

Where can I see how it was built?

Beau shared a full walkthrough of the build covering how the extension connects to Xero, how the model is hosted and run locally, the matching logic, and how to test and customise it. The article links to it through his original LinkedIn post.

Sponsored

Vinyl — the AI meeting assistant built for accounting and bookkeeping firms. Try Vinyl free.

AIM&A

Was this useful?
Thanks — noted.

Sponsored

Vinyl
Read next
Related
Listen next