Why I Started 0bs.chat
To begin with, this is mostly going to be self-reflection while watching Theo's live stream.
Most people just say don't build chat apps, and I believed the same…. Well, that was until everything started to change once my grandpa got into a motor accident. He is a retired math tutor and suffers from dementia. To help him with that, I introduced ChatGPT to him, and he instantly got hooked using the chat app, asking it various questions. It seems to have reignited the spark in him, he is solving math questions all day long and even forgets he is injured (he has diabetes so it takes longer to recover).
Well, that aside, before this I was making Minecraft servers, mods, etc. in Java and playing around a ton with image generation and Python. I even switched majors to statistics + mathematics because of this. But this wouldn't be the case if I wasn't super scared of React and CSS because I started out reading and taking advice from blogs and people that I obviously should have filtered out. Well, that left some PTSD, and I refused to even go near JavaScript (JS) for over a year.
Well, this all changed after I started using ChatGPT and Claude more and more as they improved the core AI capabilities, but honestly, the product itself was so buggy. This was followed by my grandpa getting into an accident, which led to me forcing myself and releasing my own simple chat app. While promoting it to others on Reddit (yes! Reddit), I discovered Theo's channel, and the initial t3.chat app looked awfully similar to mine and was going crazy popular, which got me all salty. But again, it all changed when he showed T3 chat with React Scan (thanks Aiden for creating this), and I was like, "Huh, now I get why, this guy is super cracked in the right places."
Again, I got blessed by ADHD gods and received an enlightenment in UX watching that stream. I started to think about the right stuff in the right places. I am still far from perfect and am still manifesting UI enlightenment, but thanks to shadcn, it's all good for now. Either way, I remember rewriting everything in TypeScript immediately after his stream (if it wasn't for Cursor and Theo, I don't think I would have). Things just started to click in. Before, I was too scared to do new stuff, partly because of all the grunt work associated with doing it and trauma, but all of this plus Theo's takes on AI unlocked a state in my mind that I had forgotten, that I can just do things.
But well, not everything is upside. I am not shilling out my university or anything, but Vancouver just doesn't have the grind mindset, partly because everyone is/has loaded as f*** parents, and it was just safer to get a job. Plus I only hear a couple of handful of people getting an actual real and fun job that they like after graduating.
Dooming and yapping aside, the way we interact with tech is changing. The future is definitely going to be vision and speech for general purpose, but AI models are not yet reliable enough currently, and the most efficient way to use and control them is using text, and always will be. The speed at which you think doesn't change much; the amount of thought varies, and typing/writing gives you enough time to conceptualize your thought. So for heavy tasks, the preferred way of doing stuff will still be text, imo. I cannot count the number of times I have thought an idea is good and tried voice vibing, only to revert that change because the initial instruction was formed due to shallow thinking.
Now let's get into the technical side of what I built: 0bs.chat
I wanted a chat app that could do everything and I mean pretty much everything I ever wanted or found useful and an app that just works, with scale in mind, which is evidently very hard. If you disagree with this, then you probably have never built stuff that a user used by choice in your entire life. Here is a feature list of what I have built and a poorly recorded n edited demo video (because I can't get my priorities straight, also shoot me a DM if you wanna help out, got bugs, etc).
https://fxtwitter.com/barre_of_lube/status/1949335384800071887
Core feature comparison: (T3 Chat is still a competitor, simply because I try to hold t3.chat as the minimum quality standard for my UI/UX) and ofc do not forget we support all popular models.
Current stack is:
- Convex (If you use anything else, don’t even bother talking to me, as you fall in either stubborn, or ignorant category, if you still think you don’t, feel free to dm me on twitter @barre_of_lube and explain why)
- TanStack Router (Why not SSR: Adds un-necessary complexity without having any real benefit in my current usecase)
- Fly.io (to deploy stdio and docker based MCP servers)
- Openrouter (I really hate it but there are really no other options for indie devs, i’d switch to litellm immediately if I had the scale, right now it makes no sense)
- Python:
- Vision Language Model (VLLM) OCR documents + audio pipeline (have tried both serverless and gpu vms), i’d love using gpu vms but again I do not have the scale, using serverless adds anywhere from 30-80% additional latency in OCR processing.
- Web Crawler for scraping docs and sites
Here's what I am currently working on:
- Browser control extension Model Context Protocol (MCP) to control your browser like OpenAI agent and Perplexity co-pilot
- Voice mode like Grok and ChatGPT (this is possibly the stack I have the most experience in)
- Enhancing React artifact and artifacts capabilities into a full-stack Convex + TanStack Router deployable app MCP, essentially Convex Chef
- Enhancing markdown artifacts and introducing /notes (a ChatGPT Canvas type) rich minimal collaborative inline markdown editor with a simple inline markdown editor for markdown artifacts with AI auto-complete and quick access AI tools. I'd actually want this right now so bad—I don't want to use Google Docs and paste stuff in 0bs.chat and back and forth. Can't wait to release this!
Socials:
- Github: https://github.com/0bs-chat/zerobs
- Twitter: https://x.com/barre_of_lube
Top comments (0)