nostr relays can respect users' privacy, just that that has not been prioritized by most relay devs

i'm not sure that basic LLM models are that resource intensive, maybe a few Gb of data of the model but walking the paths in it is not, llama, for example, 3.5gb is the model

idk about voice recognition systems, but i think they are less intensive than LLMs for resources but maybe more for compute (they do FFTs and stuff to catch phonemes)

i'm pretty sure for maybe 5 years now most phones have had all the capabilities required to do this, and definitely no problem for a max 5 year old laptop

i think the hardest part of the formula is actually privacy respecting relays... my relay #realy already does this by requiring auth for reading DMs (it's open otherwise for read) but whatever event kind you need to be gated by auth to prevent others accessing it... that would take a small amount of customization... i hadn't thought about it but it would be simple to make a configuration to set the private event kinds that must be authed to the same as the author or as tagged in the event of that kind (that is, what you sent, and what was sent to you)

WANT! 💯

I once thought about building something like it.

I would make it native apps, personal notes is very private and the browser is too general.

Nostr could be an extra optional layer to share thoughts, gather bookmarks etc. Maybe as a separate internet-connected tool, as i personally don't give internet access to note taking apps.

Tried to zap you but it failed

I’ve been brainstorming a similar app.

- Verbal dump of my day into the app.

- AI summarizes it

Each day would be a summary with ability to view the full transcript and listen to the audio, if desired

I’m planning to use nostr:nprofile1qqs8msutuusu385l6wpdzf2473d2zlh750yfayfseqwryr6mfazqvmgpz9mhxue69uhkummnw3ezuamfdejj7qghwaehxw309aex2mrp0yh8qunfd4skctnwv46z7zc8se5 for the LLM part but don’t have the speech to text yet. A small whisper model on the phone might be able to handle that.

I would love something like this! And surely running local, I don't want my crazy thoughts to be wandering in the wilderness of internetz.

Languages may be a hurdle. Some people prefer to think in their fringe mother tongue and transcriptions may be less good in those. Also whenever I try AI in such a language, the responses are not perfect.(AI seems to throw in occasionally some Croatian into Czech or Slovak).

fwiw, these are the types of apps that we are building out nostr:nprofile1qqsd79ejwuvz7v246danxqs3hgw7f2q4qrqz6x27je8er0nhfmykwzqpz4mhxue69uhhyetvv9ujuerpd46hxtnfduhszgthwden5te0vejk2ern9ehx7um5wghxyctwvshhg7tsv4ekxunfwp6quvpaka to handle. Strong e2ee privacy with the performance and UX of cloud apps.

The platform just reached “Production Ready” status. It’s running Maple AI and has another app coming online in a few weeks. Any builder who wants to play with it, hit me up. You can find out technical details of our e2ee claims on our blog.

Run local and tailscale into it to use from mobile

I’m still attached to handwriting with my apple pencil or remarkable tablet, so for me having OCR for handwriting is a must for creative flow.

The dictation feature built into iOS is local only, works offline. Not sure about the various Android variants and what they do. On iOS you could handle audio without a special model. It likely handles different languages better too.