the problem with ai note apps

every ai note app wants to chat with your notes.

"ask questions about your knowledge base!" "have a conversation with your documents!" "your notes, but with a chatbot!"

chat. i think we're cooked.

the chatbot trap

chatbots are impressive demos. you upload your notes, ask a question, get an answer. magic.

except... i don't want to have a conversation with my notes. i want to find the note and read it myself. i wrote it. i know what it says. i just can't find it.

there's a massive difference between "help me find this note" and "summarize my notes for me."

the first is useful. the second is showing off.

the real problem

when i open a notes app, i'm usually doing one of two things:

1. writing something down
2. finding something i wrote

that's it. i'm not trying to "synthesize insights" or "connect ideas across my knowledge graph." i'm trying to remember what restaurant my friend recommended.

the writing part is solved. literally every notes app handles writing fine.

the finding part is where everyone fails.

why search sucks

traditional search matches strings. you search "pet" and it finds notes containing the word "pet." if your note says "dog," tough luck.

this is fine for documents where you remember exactly what you wrote. it's useless for notes where you just dumped your thoughts and moved on.

"that meeting about the thing" is not a searchable string. but it's how we actually remember things.

what ai should actually do

ai in a notes app should make search work the way you think.

you search "pet" and find notes about dogs. you search "that meeting with sarah" and find notes about sarah that are probably meetings. you search "recipe with chicken" and find the thai curry recipe even though it doesn't say "recipe" anywhere.

this isn't magic. it's semantic search. it's been possible for years. but most apps bolt on a chatbot instead because that's what gets funding rounds.

the cloud problem

most ai features require sending your notes to a server. that's how embedding models and llms work - they run on beefy gpus in the cloud.

this is fine for some use cases. it's weird for notes. these are private thoughts. medical stuff. rants about your boss. you really want that on someone else's computer?

"but we use encryption" okay cool so you're decrypting my notes on your server to run them through the ai. very reassuring.

what jottie does differently

jottie runs ai locally. on your phone. using apple's ml frameworks.

the models aren't as powerful as gpt-4 or gemini. i'm not gonna lie about that. but they're good enough for the job. you're not asking it to write poetry. you're asking it to understand that "dog" and "pet" are related concepts.

local inference means your notes never leave your device. there's no data to leak because the data is never transmitted.

the actual value of ai in notes

- semantic search that finds what you mean
- auto-tagging that extracts people, places, dates
- connections between related notes
- maybe some summarization if you have really long notes

that's the useful stuff. not chatbots. not "ask your second brain." not "ai-powered knowledge graphs."

just help me find my notes. that's all i'm asking.

the hot take

most ai note apps are building features for demos, not for users.

chatbots look impressive at y combinator demo day. semantic search doesn't. but guess which one you'll actually use every day?

anyway. that's my rant.