IndexedDB

So if you try to read massive amounts of data, the UI freezes.

Which is why we got IndexedDB.

To build a true offline-capable application, you need a robust, asynchronous database engine running inside the browser.

If a cookie is a sticky note, IndexedDB is like having a massive personal filing cabinet right at your desk.

It's a fully searchable, client-side, NoSQL database for complex data.

Having that raw database power natively is the prerequisite for offline functionality, a web page that works without the web.

So if you're on an airplane and you lose connection, the browser intercepts any network requests.

Instead of a 404 error, it routes them to the local attach manifest.

But this is more an implementation detail.

The browser could, for whatever reason, decide to put those into blobs in an IndexedDB or a SQLite database, whatever is cheapest.

It's an implementation detail that you, as a developer, don't need to be concerned with.

You can try already some of the, like, demo implementations that I did.

So there's like implementation limits in place where you can just say this is like most browsers do it in a way that they don't look actually at the actual free disk space, but they make up, for example, 10 gigabytes free disk space to avoid fingerprinting.

so that you don't work on the actual free disk space, but on this supposed disk space.

And then you just look at how much of this is already occupied by other resources like IndexedDB, Cache API, whatever.

And you then get a share of this.

He's asking three questions.

One, how does the background upload runner likely work using IndexedDB and whatever browser APIs make this possible? Two, what's the server-side queuing and batching pattern for the Gemini calls? Three, what does backward batching actually mean, his term, and why is it the key insight? And then the meta question, how do you review code that an AI wrote when you haven't even looked at it yet? That last one is the one that's going to keep me up.

So today we're going to reverse engineer what Claude probably built for Daniel, and why the pattern matters even if your internet is fine.

Let's start with the core tension, because it's not obvious until you've lived it.

is wrong.

So we've got three layers to walk through.

client-side caching, the background thread runner using IndexedDB, server-side job management, queuing and batching those Gemini API calls, and then the layer of understanding what the AI actually built for you.

Three layers, and the third one is where the real discomfort lives.

switch it to the map.

Oh, if it's IndexedDB, you'll be like, what the heck am I looking at? Because you're having a database instead of a database.

And you're like, I can read a few things, but this looks like mumbo jumbo.

But no, if you deserialize the data out, if it's Portabuff, or actually access it to the proper APIs and all that with IndexedDB, then you can actually see what the heck is in it.

And I mean, this tool is one way of doing that.

Yes.