Show HN: Sequel – Effortlessly understand your data

bbor 5 hours ago

Looks awesome! Not the kind of thing I need, but I see the clear value prop. And the site is slick af.

Some random critiques:

1. You have a tiny typo in the “Uncover insights with AI” example (quire).

2. The docs are broken on mobile screen sizes. Not a common use case for actually reading them, but I was curious!

3. The FAQ accordions aren’t individually closable and seemingly use MouseDown instead of Click, which makes scrolling on mobile a little wonky.

4. Your (Midjourney?) images are killer. Some of the best usage I’ve seen yet for a professional site.

And finally, two actual critiques/questions on real functionality:

1. I love the most recent blog post and agree with it 100%, but was surprised to see this:

  “Instead of wrestling with complicated BI tools, users can instantly create charts, graphs, and live dashboards that help them make sense of complex data.”

That seems like a HUGE feature, with a ton of edge cases! I do now see that it’s on the home page, too. You don’t have to answer at all if it’s secret sauce, but I’d be curious to hear what library you’re using for the visualizations — presumably not direct d3?

2. Have you considered/explored paradigms that don’t require users to send you their DB credentials, and instead have some sort of local agent as a middleman? I guess the difference is arbitrary at some point (a middleman could pull my data too!), and I’m so far from an expert, but it just feels weird sharing DB login info just to get help writing queries. More curious than anything.

3. I love the privacy doc, but I’d recommend sharing any LLM providers that would get access to this data, or if you’re running it all yourself, bragging about that more explicitly! I personally don’t care about OpenAI/Anthropic/Google seeing (more of) my data, but I know a lot of people do.

4. This is all fun and games for queries, but what sort of protections are in place for INSERT, UPDATE, or, god forbid, DROP statements? I’m the biggest LLM fan around, but I think it’ll be a while before I’d let an LLM run such queries directly. Unless there’s some sort of rollback-able transaction support, and/or deterministic safety checks?

Thanks for sharing, no pressure to answer any of those lol. Very inspiring for my own work, especially as a solo dev! Keep cookin’ ;)

haxzie 2 hours ago

Wow, thank you for taking your time to go through the app!
2. Will fix the docs to work on mobile screens, this was a last moment run to setup docs, wanted to try it without any templates. Have to manually handle the responsiveness now :)
3. Makes sense, this shouldn't have caused issues on mobile. Thanks for letting me know.
4. Midjourney did a pretty good job with the images, got em done within no time! haha
---
Visualisations are done using recharts. Currently I have implemented a very limited set of charts, hopefully will add more with much more customizability.
I am planning on opensource the product soon, going ahead with an open core model, and a self hosted version along with it.
The concern about LLM providers makes sense, will add them to the docs as well.
Currently there is no hard wired instructions to not let the LLM insert/drop it. It's all as good as the user instructing it to do something and it does. There are soft instructions to not generate queries for that, but it's as good as telling a kid not to eat chocolates. I am planning to run a post analysis on the queries generated and marking them as potentially dangerous, so that the queries don't get executed automatically, but with a confirmation from the user. This should ideally be separate from the actuall LLM which generates the query.
Again, thank you so much for the valuable feedback. This helped me a lot!