An example collection
The stories, habits, and ordinary moments you don’t want to lose.
Meaning search runs on your device. First use downloads a 23 MB model plus runtime files.
Word matching is ready. Enable meaning search to find related ideas.
Start with one ordinary moment. Who was there? What do you still remember?
Fictional example memory
It doesn’t have to be a big story. The small details count.
The original text belongs to you. Save a backup, or make a simple collection to read and share.
Exports contain the text of your collection. Share only what you intend others to read.
The memory will be removed from this browser. Keep an exported backup if you may want it later.
Memory Lantern is a place to keep small stories about people, places, and moments. It was made for Hacktoberfest’s “Ghost in the Machine” theme: helping the past stay findable without inventing it.
Meaning search uses the open-weight all-MiniLM-L6-v2 model to compare your query with passages from your memories. It can connect related ideas even when the same words aren’t used. It returns original passages, never generated memories.
Results are ranked by similarity, which is not certainty or proof of relevance. It can miss a useful memory or return an unrelated one. Try word matching for names, dates, and exact details. English text works best.
The app stores your collection in this browser’s local storage. It is not encrypted, synced, or backed up automatically. Someone with access to this browser profile may be able to read it. Clearing browser data can remove it; export a JSON backup.
Loading meaning search downloads code from jsDelivr and model files from Hugging Face. These services receive ordinary download metadata. Your memories and queries are processed locally and are not uploaded by the app. Model files may remain in the browser cache. Reloading can still require internet for code/runtime files.
No accounts, analytics, microphone access, or hosted inference. Exports contain your original text; the reading copy needs no internet.
The model and Transformers.js are Apache-2.0 licensed. ONNX Runtime is MIT licensed. The original app is MIT licensed. The example collection is original fiction, clearly marked as such.
Memory Lantern