I see strong potential in this feature, however given Microsoft's history in some regards I believe this feature should be (a) fully auditable and (b) not dependent on specific hardware (like Recall requires a special NPU chip) and (c) OS independent. There's still a lot of room of improvement but ultimately we believe that the community should decide what is important (is it encryption? is it blacklisting apps?) rather than the features are dictated by a megacorp.
This first version is very much scotch-taped/hacked together of some 'AI' models for OCR/semantic embeddings and some OS API calls which get dropped in a SQLite db and screenshots stored as webp (lossless compression). We believe that there is a lot of potential to improve.
Can you elaborate on what potential you're seeing? As a typical HN user (in my case a software engineer), I have difficulty imagining in which ways it could be useful (especially on a platform like Linux for example).
Well, it's entirely optional. I understand why people wouldn't want this, but for those who do see the benefit of accessing this additional layer of memory we want to give them complete control over their data (screenshots, OCR'd text, embeddings) and auditability of the software.
8 comments
[ 3.0 ms ] story [ 23.9 ms ] threadThis is an honest question, not criticism.
FOSS is great, but the idea here is not great in the first place. FOSS or not.