Otto Engineer

@remango wow. First of all, incredibly polished launch with a slick demo and clear pitch. I also learned about Web Containers, which makes so much sense...I will try your product out. Congratulations.

@aris_nakos Thank you! 🙏🏻

@anandreddyks It is currently stateless in the sense you describe. High on my list is to allow importing a git repo or folder as a starting point, then download the result once Otto is finished. This ties into the super (not so) secret plan to have Otto orchestrate multiple, child Otto agents in order to build entire apps 😎

@jopie Yes! This is high on my list because this is the #1 thing that I myself want to use it for. There will definitely be size limitations, but I think there's still massive value in using it for mini to small to (maybe one day) medium repos 🙂

@remango Awesome, I think that would immensely increase the value it can bring. Does the size limit come from the LLM context or something else?

@jopie I agree! And yes, it's primarily a limitation of the LLM context length, but also cost comes into play since if we were to simply feed the entire chat history to the LLM on each request, the number of tokens used increases drastically. I have some techniques for vastly minimizing the number of tokens used to work on a *new* codebase written by Otto, but this only works if it's an AI-first codebase (did I just make up that term? 😝), not an existing one where we don't control how the code is structured

@sandradjajic Thank you, Sandra!!

@cwbuilds1 Exactly 😂 And yes, running in the browser via web containers means: 1) zero setup, super easy to jump into a chat 2) code is executed safely in isolation and file changes don't affect your real file system

@cwbuilds1 So what I've found Otto really excels at at the moment is a small yet complex/hairy TypeScript problem, which is what inspired me to build it in the first place (so I optimized it for this, intentionally or unintentionally). I would often get hung up trying to build a complex type (say something with type inference and generics and mapped types), and I would try to use tools like ChatGPT to help, but I would get code that doesn't work at all and have to go through many rounds of iteration. The key difference that makes Otto able to give you *actually* working code, is that it runs and tests its own code, looks at any errors, and improves the code on autonomously. I think my favorite example, though it's kind of a smaller scope one, is that snake case TS utility type that I have in my Loom demo (and also on the landing page) 🙂 It's not that crazy of an example, but it's super *useful* for me in my day to day workflow as an engineer.

@abhilash_chowdhary Thank you! And by the way, the team is just me!

@sabari_nathan5 Thank you!