Show HN: Nova nine months later – updates after a year working with GPT
asknova.ai
Short version : a big piece of feedback was that it was hard to use, so I’ve completed a massive usability and onboarding pass, culminating in this overview video capturing the workflows and features available and I’d love for you to check it out : https://www.youtube.com/watch?v=Lq8lVwYMdJM
Long version : my initial share was an early prototype of nova, a customisable ai assistant, you can see the thread here: https://news.ycombinator.com/item?id=36651557
Since then
- got some great feedback and burst of users - and everything held together!
- got structured json responses running internal commands, start of independent ‘agents’
- created ‘loadouts’ which are sets of cartridges as agents with their own workspaces
I have been able to use it for work and side projects and an interface for configuring agents has come in really handy. I have felt a bit discouraged as similar work came out, but wanted to do my idea justice, so have kept at it.
Recent progress
- added past conversation retrieval and recall grouped by loadout
- switched over to open AI function calling and improved the async workflows
- added web search, image generation, video editing, and better note handling
- plus lots of refactoring, bug squashing and trying new stuff like voice
Some numbers
- got up to $500 + spent on credits with openAI per month (whoops)
- accidentally spent about $4k one month, (don’t tell my wife)
- lots just using guest nova profile, including lots of people in china (你好，中国朋友们!)
- had a few hundred different unique sign ups, and a few regular users
- I’ve typed something like a million characters personally, I use it every day
Current stack
- still react frond end, pythong back end, quart + prisma -> planet scale, big area for optimisation there, plus some db handling I’d like to tidy up
- I am starting to do more compute heavy stuff, eg video editing, and looking at breaking those up into discrete services, probably migrating to AWS (from railway), so I can pass jobs to the right hardware
Where to next
So I basically committed myself to finalising what I’d call ‘my current sprint’ - bug free, better onboarding and usability, solid deployment to clients. I’m declaring sprint complete as of this post (about three months behind schedule).
However I’m still pretty obsessed, my next set of goals are:
- much better knowledge management and retrieval : go from a patchwork of file handling and retrieval to a unified system and interpreter
- expose agents to each other so they can both query and retrieve based on other agents threads, but also action and make requests.
- let users add their own API key, expose nova api to more clients and add other models and ideally do desktop version
One thing I’ve considered is open sourcing at least a version of the client and server as packages. I haven’t done anything open source, but could see it as an interesting challenge.
Anyway, I hope you get a chance to try out the updates, you can test them out at https://asknova.ai/
I’d love to hear about your onboarding experience, especially if you feel a sense of continuity from the guest to logged in use. If you have ideas on how to utilise it or feature requests I'd love to hear from you.
If you have any feedback, or want to chat, hit me @samueltates on twitter or me@samueltate.com
thanks HN and have a great year!