Tresor.co or cloud.near.ai could be a useful inference provider for this kind of project as these providers run inference inside enclaves which verifiably cannot peek at your data even as the hypervisor. Both are open source. Tresor.co is EU based so I'm thinking of using tresor.co for one of my projects as I trust EU regulation more for this sort of thing.
I like the idea behind this a lot and feel the same way about my notes. I also built my own app for it and have a very similar feature on the roadmap. Cool stuff!
I just tried the android application, really like the privacy first approach. Would you consider adding BYOK so that a user could have complete control over where inference happens? I realize it might not fit your architecture, but it seems like a neat optional setting. Either way, neat project.
I did consider it at one point the problem is that the prompt and instruction following for formating and extracting tags, nouns etc and the json returns are unreliable if you switch models and especially if it's an older model. right now i'm using Gemma 4 and works smoothly so it would need to be BYOK for gemma only which is doable. If more people ask I could do it sure.
the best would be on device inference only but we need maybe 6 months for a small model to match capabilities yet. I wanted to do it like that but even my pixel 10 suffered.
Same reason here: I got tired of my stuff living on someone else's server, so I ended up building my own local memory thing. Not a developer at all, just stubborn.
Two honest questions, since you built it for yourself first:
1. Are the entries plain files I can open and read without your app, or a database? For me that ended up being the whole point. If I can't read it without you, it's not really mine.
2. You say everything is on device, but inference goes to a zero-retention endpoint. Have you looked at running a small local model for the AI part? "Private" and "my text leaves the machine" are hard to hold together in one sentence.
Not a criticism, I hit the same wall myself. Nice project.
1) the app is using a database stored on your phone. you can export it all in json or llm format. the one is for export/import and the other to feed an LLM with your data to chat more
2) yes I tried it and it becomes so slow it's unworkable right now but in a future version totally possible. Gemma 4 needs an iphone 17 pro or similar to run in a normal timeframe and still it's slow and also the whisper-large is also very slow. it would also need to download some gigs of data. but that was actually the first version I built :)
Asking an AI how you feel about a person or the highlight of a trip is such an offloading of basic human identity.
Tresor.co or cloud.near.ai could be a useful inference provider for this kind of project as these providers run inference inside enclaves which verifiably cannot peek at your data even as the hypervisor. Both are open source. Tresor.co is EU based so I'm thinking of using tresor.co for one of my projects as I trust EU regulation more for this sort of thing.
I like the idea behind this a lot and feel the same way about my notes. I also built my own app for it and have a very similar feature on the roadmap. Cool stuff!
I just tried the android application, really like the privacy first approach. Would you consider adding BYOK so that a user could have complete control over where inference happens? I realize it might not fit your architecture, but it seems like a neat optional setting. Either way, neat project.
I did consider it at one point the problem is that the prompt and instruction following for formating and extracting tags, nouns etc and the json returns are unreliable if you switch models and especially if it's an older model. right now i'm using Gemma 4 and works smoothly so it would need to be BYOK for gemma only which is doable. If more people ask I could do it sure.
I would not use a personal journaling app that didn't give me that type of control, so here's one more asking :)
the best would be on device inference only but we need maybe 6 months for a small model to match capabilities yet. I wanted to do it like that but even my pixel 10 suffered.
I downloaded it today. Pretty cool
Great,maybe a Great Work!
Same reason here: I got tired of my stuff living on someone else's server, so I ended up building my own local memory thing. Not a developer at all, just stubborn.
Two honest questions, since you built it for yourself first:
1. Are the entries plain files I can open and read without your app, or a database? For me that ended up being the whole point. If I can't read it without you, it's not really mine.
2. You say everything is on device, but inference goes to a zero-retention endpoint. Have you looked at running a small local model for the AI part? "Private" and "my text leaves the machine" are hard to hold together in one sentence.
Not a criticism, I hit the same wall myself. Nice project.
1) the app is using a database stored on your phone. you can export it all in json or llm format. the one is for export/import and the other to feed an LLM with your data to chat more
2) yes I tried it and it becomes so slow it's unworkable right now but in a future version totally possible. Gemma 4 needs an iphone 17 pro or similar to run in a normal timeframe and still it's slow and also the whisper-large is also very slow. it would also need to download some gigs of data. but that was actually the first version I built :)