Personal Voice Memory
Speak a thought.
Find it months later.
ScribeSwift records your voice, transcribes automatically on your own server, and makes memories searchable by meaning — not keywords.
Self-hosted · Open source · Private
The thought arrives at the wrong moment.
You're running. You're driving. You're three seconds before sleep. It's the clearest the idea has ever been, and you have about thirty seconds before something else takes its place.
ScribeSwift is what you reach for then.
Capture now. Retrieve later.
Four steps from thought to found.
Tap Record
The app opens. Tap once. The microphone starts. No loading screen, no decisions.
Speak
Say the thought clearly — or not. Whisper works too. Tap Stop when you're done.
It transcribes
Your server runs Faster-Whisper on the audio. Words appear within seconds.
Find it later
Search by meaning, not exact words. Related memories surface automatically.
One tap.
No preamble.
Most note-taking apps make you decide something before you start. What folder? What project? What kind of note is this?
ScribeSwift does not. The record button starts immediately, with no loading screen and no modal asking you to categorize your thought before you have finished having it.
Open the app. Tap Record. Speak. Tap Stop. That is the entire interface.
Search by what you meant.
Not what you said.
Two months after a walk where you had an idea about pricing, you remember it as "the pricing thing from the walk." Not the exact words.
Type that description. ScribeSwift finds the recording.
It works because every transcript is embedded as a semantic vector — a mathematical representation of meaning rather than a bag of words. Similarity is computed by direction, not exact match.
Related memories also surface automatically. Open any recording and see the past thoughts that share its meaning — without searching.
Privacy
Your recordings never
leave your server.
ScribeSwift processes everything — transcription, embeddings, analysis — on hardware you control. There is no cloud API in the pipeline.
This is not a privacy policy. It is an architecture.
The audio files sit in a directory on your server. The transcripts live in JSON files on your server. The semantic vectors are computed by a model you downloaded, running on a machine you can name.
Nothing is sent to Anthropic, OpenAI, Google, or anyone else. Not as a current practice. As a structural fact.
Your thoughts don't need
permission from a server.
Recording, playback, and keyword search all work without a network connection. The app is installed on your device as a Progressive Web App — its full interface is cached locally.
Transcription and semantic search need your server. But capture never waits. The thought is safe the moment you tap Stop — wherever you are.
Full capabilities
- Record audio
- Transcription
- Semantic search
- Sync & pull
- Playback
Defined capable state
- Record audio
- Keyword search
- Playback
- Transcription
- Semantic search
The AI works silently.
There is no chat interface. No "ask your notes a question." No AI-generated summaries of things you already know.
When you sync, ScribeSwift transcribes the audio, generates a title, identifies the type of thought (task, idea, reflection, business, technical, personal), extracts action items, and creates a semantic embedding for retrieval. All of this happens on your server. None of it requires your attention.
The AI is not the product. Your memories are.
Self-hosted.
Private by design.
Yours to run.
ScribeSwift runs on a server you control — a VPS, a home server, or a spare machine. Setup takes about an hour following the deployment guide. After that, it runs silently and maintains itself.
This is the trade: a bit of setup in exchange for unconditional privacy. For many people, that trade is worth everything.
Self-hosted first · No third-party account · Your data stays on your server