A local-first personal database you can extend with apps and manage with AI
1 points by pscanf
1 points by pscanf
Hello all. I'm sharing this project (Superego, see on GitHub) I've been working on for way too long at this point, hoping to get some feedback and some early adopters.
As the title says, Superego is:
Example:
(You can try this exact use case at https://demo.superego.dev, no login required.)
Some technical bits that I find interesting to discuss:
Storage: saves everything in a single SQLite file. I'm trying to make it sync via Dropbox (or other file-syncing services), but I haven't cracked this yet. I tried to use git as a storage backend, but I couldn't get acceptable perfs out of it (once the db size gets non-trivial).
Security: uses iframes and QuickJS as sandboxes to safely run AI-generated code. There's a lot of talk about this, and many new products, but I've found these two "relatively ancient" technologies to solve the problem quite well in my (maybe very simple) use case.
UX: the app is completely malleable, which makes it difficult to approach for non-technical users (the problem Lotus Notes had ~30 years ago). AI makes it a lot easier, but I think we're still at level 1 of what can be done from a UX point of view, and it's really exciting to explore the possibilities.
Let me know what you all think!
The flagging is getting a bit out of hand. Clearly, this post is neither off-topic nor spam, so just hide it and move on if it gets your knickers in a twist.
Thanks.
I'm also surprised by the spam reports. I'm relatively new to lobste.rs, but I thought showing one's work wasn't inherently against etiquette and spammy behavior.
Maybe I should have just avoided putting AI in the title. Definitely a polarizing subject.
If it wasn't a slopware, I would have liked it.
I have been using AI for coding, but by my definition of slopware (= entirely vibe-coded with little oversight) it's definitely not that.
On the contrary, as I say in the README, I've always been quite picky about code quality, so I'd find it impossible to accept LLM-generated code without a thorough review. (Which might actually make it less efficient as a process. Jury's still out on that.)
But fair enough if you consider any AI-assisted work slopware. I myself feel like using AI still definitely has some impact on the end result, so I can understand avoiding software made with it.
But if your definition aligns more with mine, can I ask, what gave you the impression of it being slopware? (I'd obviously like to combat that impression!)