{"id":50441490,"title":"Show HN: Valin – your private memory for everything you save","points":2,"user":"dradu","time":1794323415,"time_ago":"3 hours ago","type":"link","content":"

Hi HN,

I build Valin because my wife needed a way to search across all the saves and all the information she came across online - from recipes, books, children activities, vacation recommendations. She didn't want another AI subscription or another account to manage.

Valin was born - an application that allows users to share to it (almost) any kind of contents from applications (videos, audio, links, papers, reels. etc) and it goes in and using on-device AI transcribes audio, extract key frames of the video, extracts text from the frames and shows Key Takeaways and Summary. This allow users to get an at-a-glance overview of what the save was about and it allows the application to cluster and to find similar saves on the device - to generate new collections or to automatically (if enabled) assign saves to existing collections. And in the end it allows the system to support a chat-like interface where users can quickly get the information they need, not just the save.

I vibe coded it in the last 5 months and I'm pretty happy how it turned out so I wanted to share the TestFlight beta to see if its interesting or useful to other folks.

Features

- fully on-device: no cloud services, no AI subscription, no account needed.

- supports videos, audio, pdf, images

- supports applications: FB, Instagram, Substack, X, Reddit, TikTok, Threads, Voice Memos

- supports iPhone (including Duo) and iPad

- iCloud sync (optional) to allow cross device sync and processing

Limitations:

- it can only work with saves that are available or can be read without a subscription (no logins)

- it only processes in the foreground or background if you approve it when suggested.

Requirements:

- iOS only

- iPhone 15 pro or newer

- iPad with M1 or newer, or the iPad mini with A17 Pro.

Models used:

- Gemma 4 E2B - for summarization, chat and general information extraction

- Embeddings Gemma 300m - for clustering, collection generation and semantic search

- Whisper - for transcribing

What I'm proud of - there are over 10 TLA+ models which I used to model everything from the DAG that processes the saves, to iCloud sync, to GPU/CPU processing of AI requests (since iOS doesn't allow background GPU processing on the iPhones).

It's a beta right now on TestFlight - https://testflight.apple.com/join/gjJQQrqk\nWould love to hear feedback and suggestions.

Daniel","url":"https://www.valin.ai/","domain":"valin.ai","comments":[],"comments_count":0}