Ofrenda — a memory altar that never leaves your device

01 Bring a memory

Audio never leaves this tab. If you'd rather type, that works too — the writer model still does the writing.

00:00

Ask them one question and stay quiet: “What do you want us to remember about you?”

⤒ Drop an audio file, or click to choose m4a, mp3, wav, ogg — decoded locally in the browser

A short memo from Rosa, recorded in 2019. Loaded from a bundled text file — no network request.

Model settings

Speech model Spoken language

Weights are fetched once from the Hugging Face CDN, then cached by your browser. After that the whole app works with the network off.

Fix anything the model misheard. This is your transcript — edit it freely.

Three details turn a transcript into something worth keeping.

Their name Who they are to you One thing to keep

Card language Title on the card Dates (optional)

Let the local writer model compose the text (recommended)

Download it, print it, or hang it. The link in your address bar carries the whole card — it opens anywhere, no account needed.