Antonio Pinto·December 9, 2025·3 min read

First, the Boring Important Part

You can't create an AI email platform without having first a strong email client

At Onbox, we're building an AI assistant for ops teams that lives in your mailbox. It learns your processes, connects to your tools (ERP, CRM), and helps you close cases faster.

But even with AI handling routine tasks, you'll still review lots of decision and you'll still be browsing a lot.

That's why making a fast email client is important. A slow, clunky interface doesn't just annoy you—it kills the whole AI experience.

So before building the smart stuff, we built a strong foundation, optimizing for speed and reliability.

Instant response times ⚡️

When you're moving through emails, every tiny delay adds up. A loading spinner here, a flickering screen there—these micro-frustrations drain your focus.

The problem? Most email clients fetch data from servers. Even with a fast connection, that round trip takes time.

That's why we went local-first. Your emails live in a database that runs directly in your browser (yes, a real PostgreSQL database, thanks to PGLite). When you open a thread, it loads instantly from local storage. When your connection drops, you keep working.

The difference is very noticeable. Opening threads feels instant. Switching between conversations has no lag.

Search that is fast AND powerful 🔎

To build a great search, we needed powerful indexing and lookup capabilities. Doing local wasn't giving us good results.

That's why we built the search infrastructure on OpenSearch, which gives us both speed and complex querying.

It searches across sender names, email addresses, and content. It supports fuzzy matching for typos, complex filters for AI queries, and advanced ranking techniques we can now tune.

The foundation is solid. Now we're tweaking relevance and ranking based on real usage. Getting search to feel truly magical takes iteration, but we have the right platform to do it now.

Emails that look right ✉️

This one feels sooo boring, but it's surprisingly hard and important.

Email HTML is a mess. Some emails have nested quoted replies going back months. Some use white backgrounds that blind you in dark mode. Some contain tracking pixels and unsafe scripts. And different email clients send wildly different HTML.

We had to build preprocessing to clean all of this: identify and collapse quoted content, strip white backgrounds for dark mode, remove unsafe elements, normalize the HTML so it renders consistently.

The payoff? Your emails look clean in Onbox, respecting most of the time the theme of the app, whether they came from Gmail, Outlook, or some ancient CRM system.

And because ops people juggle lots of attachments, we show visual previews. Now you can spot that signed contract among 10 PDFs at a glance, instead of clicking each one.

Never losing emails 🧱

You can't miss emails.

Email sync is fragile—APIs fail, rate limits hit, webhooks get dropped.

That's why we built monitoring that catches problems before you do. If threads go missing, we detect it. If sync stalls, we retry. At every critical step, we have fallbacks.

The result: emails flow reliably into Onbox, and you can trust what you see.


Now we have a foundation that works. Fast, reliable, clean.

And that means we can build the AI features on top of something solid.

See if this inbox is for you

Be among the first teams to try Onbox on real cases.

join 100+ enthusiasts