SQLite

Well, what we got to do is do this right.

And so were you already imagining what kind of success looked like, like as in, everyone doing any sort of Ananti stats, not quite SQLite, will be using DocDB.

Do you have that picture in the beginning? Like, well, this is where it's going to be.

Wow.

It's all safe.

But it didn't work to have two processes operate on the same database file.

This works on SQLite because SQLite cheats and just locks and then the other side busy loops until the lock is lifted and then it does its change and the other side busy loops.

And they have zero, SQLite has zero in-memory state about the database.

They make it really easy to deploy.

They also went like full... like almost Next.js and have a, like you could use it with like TenSecStart or any of these existing frameworks, but they also built their own little meta framework that works super well with things like routing, integration with database, you know, like basically they took every single little Cloudflare integration, object storage, cron jobs, key value, cache, emails, queues, AI gateways, SQLite, all of that.

And they made adapters for each of those so that you can use it directly in.

They kind of made, like we've been asking forever, like where's the Laravel of the JavaScript world? They kind of made that.

They have a table in their repo here.

that talks about the different things that they can analyze.

So you've got like Rust binaries, Wasm, SQLite databases, Electron apps.

And basically for someone that doesn't know anything about this space, it's basically just a guide.

You could also use JSONL, which is an append only file system.

But for real long running production agents, the developers strongly recommend using SQLite.

And the sources mentioned this utilizes right ahead logging or wall mode.

I want to stop you there for a second because whenever I hear right ahead logging, I usually think of traditional database recovery for like e-commerce sites or banking apps, not AI brains.

And because the LLM is dynamic, it can read that error receipt, evaluate the partial output, and logically decide on the next best step.

It uses reasoning to recover rather than relying on a rigid workflow system that would just throw a fatal exception and break your production server.

Because this agent's brain is effectively trapped safely in a SQLite database, completely aware of its state and immune to memory loss, it unlocks a massive shift in how we work.

We don't need to babysit the execution anymore.

You can pick up where you left off.

Yeah, and they do have two different backends that I can see, SQLite and Postgres, which is neat.

They also support standard schema for your type validations and stuff, and schemas, schema libraries, OpenTelemetry, Sentry, and Prometheus metrics.

And if you're considering it, but you're wondering what is the difference between Open Workflow versus some other tools, they do have some comparison.