How I Stopped Manually Chasing Invoices and Let AI Do It (with Nanobrowser)

Many of us — accountants and business owners alike — are familiar with the process of collecting invoices for every expense in order to maintain clean and complete bookkeeping records. Frankly, it's a task I hate.

Sometimes, this process is simple. You might forward an invoice email, and your bookkeeping software automatically scans the fields and does the rest with minimal intervention. That's fantastic when it works.

However, the reality isn't always so straightforward.

The Challenge of Manual Invoice Retrieval

Numerous online service providers for digital goods—like cloud storage, hosting, and various other services—don't always integrate seamlessly with automated bookkeeping processes. They often provide user-friendly dashboards where you can view and download your invoices. While this is convenient for occasional access, it becomes a daunting task if you need to automate the collection of dozens of invoices from multiple providers.

But here's some good news:

AI can help with that too. And it’s open source.

But let's take a step back.

While technology for similar automation tasks has been available for some time, it often came with limitations:

• Technical Complexity: Some powerful models and tools were difficult to run, typically requiring developer skills (e.g., managing Docker containers, setting up Streamlit apps).
• Some black box solutions offer models that work perfectly for proof of concept, but fail in real-world scenarios.
• When models fail in the middle of a long task, they simple start over, which can become expensive if they fail often. Instead, you might want to intervene if necessary and continue from there (human-in-the-loop).
• High Costs: Certain solutions, like earlier OpenAI offerings, had significant monthly subscription fees (e.g., $200), making them unaffordable for many.
• Tool Limitations: Other browser automation tools (like Playwright and similar solutions) often had their own drawbacks, such as requiring their own specific browser instances or operating as "black boxes" with limited transparency.

Introducing Nanobrowser

Recently, an open-source project called Nanobrowser emerged and particularly caught my attention.

Nanobrowser is a fully open-source Chrome extension. It supports a wide range of AI models currently available, including reasoning models, to accomplish tasks from start to finish directly within your browser.

This means no separate application instances or complex sandboxing is required (though for added caution, you can always use an isolated browser instance like Chromium if you prefer).