Dump Memory - We fix your memory
by•
Dump is your private second brain. Capture notes, photos, voice memos, documents, screenshots, and X bookmarks in one place. No folders, tags, or filing. Dump enriches information with web context and lets you find anything using on-device semantic search. Everything stays on your device and personal iCloud with zero cloud storage, zero tracking, and zero compromise. Just dump, remember, and instantly recall what matters.
Replies
how does the semantic search actually work on-device without sending data out, especially for the web enrichment part?
Dump Memory
@gecit_gize13066 It’s not fully on-device today.
Search/ranking happens locally: memories and enriched web items are stored in a local SQLite index, and the app searches that index on-device.
But semantic embeddings are generated through OpenRouter (text-embedding-3-small), so memory chunks and queries are sent out for embedding. Web enrichment also uses external web/search tools when requested, then the enriched result can be saved and searched locally afterward. All data store only on local device.
The no-folder approach actually works once you try the semantic search, found a random voice memo from last week just by typing what it was about. Love that nothing leaves my device.
the on-device semantic search without any cloud storage is a really thoughtful choice, especially for a tool meant to hold your most personal notes.
How does the on-device semantic search actually perform when you've dumped thousands of items?
Dump Memory
@soneragah57306 It scales to thousands of saved items using a local hybrid index, not a full vector database. We cache embeddings, only re-embed changed content, and combine semantic and keyword search for fast, accurate results.
The embedding model isn’t updated on-device. Embeddings are generated via OpenRouter using text-embedding-3-small, while the local index updates as your memories change.
The semantic search actually works surprisingly well, found a voice memo from last week just by typing a vague idea. Love that nothing leaves my device, feels like the notes app I always wanted.
Dumping everything without sorting feels weirdly freeing, and the semantic search actually pulls up that random recipe I screenshotted weeks ago. Love that it stays on device.
How does the on-device semantic search actually perform when you have thousands of dumped items, and does it work offline once the initial setup is done?
Dump Memory
@hamzagayricztz It scales to thousands of saved items using a local hybrid index, not a full vector database. We cache embeddings, only re-embed changed content, and combine semantic and keyword search for fast, accurate results.
The embedding model isn’t updated on-device. Embeddings are generated via OpenRouter using text-embedding-3-small, while the local index updates as your memories change.
The dump it and forget it approach actually works, dumped a random mix of screenshots and voice notes and found them all later just by typing what I vaguely remembered. Love that nothing leaves my device.
The no-folders approach is such a relief after years of fighting with tags that never quite fit. On-device semantic search actually sounds like it delivers on the promise instead of just adding another layer of friction.
How does the on-device semantic search actually perform once you've dumped a few thousand items, does it still feel instant or does it start to crawl?