Local-first recording, transcription and meeting workspace for Windows.
Memento records meetings, interviews, lectures and dictation from any combination of microphones, system audio and individual applications, keeps every source as its own synchronized track, transcribes on your PC, identifies speakers, and turns the result into documents you can trust. Everything stays on your computer unless you choose to export it. External AI is optional and off by default.
Status: 2.0.0 adds video (a display, a window or a camera beside the audio), a live transcript while recording, speakers remembered by their voice, suggested chapters, line selection and filters in Review, Tesseract for agenda photos, a follow-up email and a full preview of what would be sent, the tray and starting with Windows. 1.x recordings, settings and models carry over unchanged. What was verified and what comes next is in docs/ROADMAP.md; the user guide explains everything step by step.
- Open the latest release.
- Download
MementoApp-win-Setup.exe. - Run it. Memento installs for the current user and needs no administrator rights. It checks for new versions, downloads them in the background and installs them when you restart it (see Updates).
Requirements: Windows 11 (Windows 10 version 2004 or later should also work; screen recording there always shows Windows' yellow capture border, which only Windows 11 lets Memento turn off), 64-bit. A GPU speeds up transcription, the local AI model and video encoding but is not required.
The installer is not yet code-signed, so Windows SmartScreen may show a warning the first time. Choose "More info" → "Run anyway" if you downloaded it from this repository's Releases page. To check a download, compare its SHA-256 (Get-FileHash MementoApp-win-Setup.exe) with SHA256SUMS.txt on the release page.
Version 2.0.0:
- Record from microphones, everything the PC plays, or one application at a time, with each source saved as its own synchronized track and checkpointed to disk every 30 seconds. Pause, mark highlights with notes, and turn sources on or off mid-recording.
- Record video (2.0) of a display, one window or a camera beside the audio, as H.264 MP4 on the graphics card's encoder or the processor, with the camera as a picture-in-picture inset you can move later. Video is a track in step with the sound: a crash keeps it up to the last second or so, and if it stops the audio carries on. Watch it in Review in step with the transcript, export it as one file or two, or remove it later and keep the audio.
- See a live transcript (2.0, off by default) while you record: a rough draft from the Small model, a few seconds behind, never saved, replaced by the full transcript after you stop.
- Never lose a recording: a crash or power cut is repaired at the next start, and the recovery dialog says what was saved and what may be missing. A device that disappears ends only its own track; a full disk or a missing library drive stops cleanly with everything kept.
- Store every recording as lossless FLAC with a mix, a waveform and SHA-256 hashes, or choose smaller AAC/MP3 files that Memento converts and verifies after saving. Keep only the mix (2.0, off by default) removes each source's own track after processing to save space.
- Transcribe on your PC (Whisper, on the graphics card or the processor) with word-level timings that follow the voice, and confidence marks. Nothing is uploaded. Models are downloaded in Settings and checked against their published SHA-256 before use.
- Tell speakers apart on your PC: rename a speaker once and every line follows, move lines, merge or add speakers, or say how many people spoke and reduce to that number. Remember speakers by voice (2.0, off by default) suggests a name you gave before when the same voice speaks again; you confirm it.
- Review with a player, waveform, seeking, Skip silences, chapters, highlights, details and a full processing history, next to a transcript that follows the playhead: click a line's words to correct it, search, filter by speaker, highlights, uncertain or corrected lines and chapter, select several lines (2.0) to give them a speaker, highlight or copy them, right-click a line for its menu, undo any change, and open or restore earlier versions from History. Suggested chapters (2.0) mark where the subject changes, worked out on your PC.
- Copy the transcript, filtered lines or a document to the clipboard, formatted for Word and Outlook or as Markdown.
- Find recordings in the Library by title, people and words in their transcripts, filter by type, and delete them with a clear confirmation.
- Import an agenda from Word, PDF, Excel, CSV, Markdown, text or a photo of a printed page, parsed on your PC with Windows' text recognition or Tesseract (2.0, installable, marks the words it was unsure of), with unsure items marked and explained; tick items off while recording.
- Attach files to a recording (the agenda's original is kept automatically); documents, images and media open with their app, anything else opens its folder.
- Import existing audio or video (the sound of a video) as a recording that is transcribed like any other; the original file is left as it is.
- Create documents: minutes, summaries, action items, decisions, quotes, the full transcript and more, laid out in a visual builder from templates you can change, styled with presets or your own styles, and edited, regenerated and versioned in a document viewer. Every statement links back to its moment in the recording and is checked against the transcript; "How this was made" lists what was used and sent. Three providers: a local model that runs on your PC and sends nothing (good for a first draft with citations and review), or Claude or ChatGPT with your own key, off until you allow external AI, with a full read-only sheet (2.0) that shows every request word for word before anything is sent. A follow-up email section (2.0) writes a recap with the decisions and action items, in the tone you choose, ready to copy into your mail program.
- Export exactly the parts you want: the mix or each track as FLAC, WAV or MP3, the video as one MP4 with the camera inset or as separate files (2.0), the transcript as JSON, Markdown, text or SRT (with or without timestamps and speakers), documents as Word, PDF or Markdown, the details and the attachments, with a
manifest.jsonof sizes and SHA-256 fingerprints. - Update itself from this repository's GitHub releases: checked at start and once a day (never while recording or processing), downloaded in the background, installed when you restart. You can turn it off in Settings › General › Updates.
- Move the library to another folder or drive, with every file checked before the old copy is removed; and complete Settings, including storage use, encrypted AI keys and About with the licenses of every component.
- Stay out of the way (2.0): keep running in the notification area when the window is closed, with a recording in progress shown on the icon, and start with Windows (both off by default).
Next (see the roadmap): selecting recordings in the Library for bulk export or delete, context menus there, layouts below 1024 pixels, more AI providers and calendar integration.
Prerequisites: .NET 8 SDK, Node.js LTS, Windows 10/11.
git clone https://github.com/silent-diffusion/Memento.git
cd Memento
cd ui && npm ci && npm run build && cd ..
dotnet build Memento.sln
dotnet run --project src/Memento.AppRun the tests:
dotnet test Memento.sln
cd ui && npm test| Document | Purpose |
|---|---|
| docs/PRODUCT-SPEC.md | What the application must do |
| docs/ARCHITECTURE.md | How it is built: stack, storage, pipelines, reliability rules |
| docs/ROADMAP.md | Milestones and acceptance criteria |
| design/DESIGN.md | Design handoff: tokens, components, every screen |
| docs/USER-GUIDE.md | How to use Memento: recording, review, documents, export, settings, backups, recovery |
| docs/RELEASING.md | The release checklist |
| docs/THIRD-PARTY.md | Bundled components and their licenses |
| docs/SECURITY.md | Threat model, what Memento promises, how to report a vulnerability; audits in docs/audits/ |
Recordings, transcripts and documents are stored in a library folder on your PC (by default under your local application data, never in a cloud-synced folder). Memento makes no network connections except:
- the update check, the one connection Memento makes by itself: it asks this repository's GitHub releases for a newer version when it starts and once a day (never while recording or processing) and downloads it. The check sends nothing about you or your recordings. Turn it off in Settings › General › Updates ("Check now" still works);
- downloading a transcription, speaker, text-recognition or local AI model you choose to install;
- sending the inputs you tick to Claude or ChatGPT, only after you allow external AI services and supply your own key (the local model sends nothing).
Audio and video are never sent anywhere: video is never transcribed, analysed or given to any AI, and the screen, window and camera recordings stay in the library folder like the audio. The live transcript and suggested chapters are made on your PC. With Remember speakers by voice on (off by default), the voice signatures (a short list of numbers per named person, never sound) are kept on this PC in the library's voices folder, are never sent, and are forgotten from Settings › Speakers or when the recording they came from is deleted. There is no account and no telemetry. The Microsoft Edge WebView2 runtime that draws the window has its own background connections to Microsoft (runtime updates and configuration) that Memento cannot turn off; they carry nothing from your recordings, because the window never loads anything from the internet. Report security problems privately through GitHub Security Advisories (docs/SECURITY.md).