Skip to content

Privacy

ArchiveMind is built to be local-first: indexing and search are designed to run entirely on your own computer.

What ArchiveMind does not do

  • It does not upload your files. Indexing a file means reading it locally and storing what was learned in your local catalogue. Original files are not sent anywhere merely because they were indexed.
  • It does not move, rename, or modify your files. Indexing only reads.
  • It has no telemetry. The private-beta builds contain no analytics, no usage tracking, no phone-home, and no auto-update mechanism.
  • It has no account or cloud service. There is nothing to sign in to.

A claim we deliberately do not make

We avoid the blanket phrase "nothing ever leaves your computer" until every code path — including future update checks or model downloads, should they ever exist — has been verified for each release. What we can say about the current packaged beta builds: the visual-search model ships inside the app, indexing and search run locally, and there is no telemetry, no analytics, and no auto-update. One documented exception in earlier beta installers: the first launch of the macOS .command-based beta used the internet once to install its Python dependencies; the packaged app builds bundle their dependencies instead.

What ArchiveMind stores (derived data)

To make search work, ArchiveMind keeps a catalogue of derived data about your files:

  • file paths and filenames;
  • sizes, dates, and file types;
  • content fingerprints (hashes) used for duplicate detection;
  • metadata read from files (for example photo dates, camera details, audio tags);
  • text extracted from documents, up to an extraction limit;
  • thumbnails and media previews;
  • visual-search index data (embeddings) for photos, where enabled;
  • repository (drive) records and their availability states;
  • operational logs (see below).

Derived data is still personal data

Paths, filenames, extracted text, and thumbnails can themselves reveal a lot about you. Treat the catalogue like you treat the files it describes:

  • Protect the catalogue database and the application-data folder. They live on your machine, under your user account — full-disk encryption and a strong login password protect them the same way they protect your files.
  • Be deliberate when sharing. Do not send your catalogue, logs, or application-data folder to anyone you would not send the files themselves. When reporting a problem, use the sanitized diagnostics export instead — see Reporting a problem.

Where the data lives on your platform: Windows · macOS.

Logs

ArchiveMind writes bounded, rotating operational logs (what happened, when, with which counts and statuses). Logs never contain file contents, extracted text, or thumbnails — but they may legitimately contain file paths, which is why the diagnostics bundle redacts paths by default.

Local processing vs. optional downloads

Everything described above — scanning, hashing, extraction, thumbnails, visual search — is local processing. Two kinds of optional download exist, and both are things you choose, not things ArchiveMind does behind your back:

  • Installing optional tools (such as ffmpeg for media previews) means downloading those tools from their own projects using your package manager. That is a download you initiate to install software; it is not ArchiveMind transmitting your data.
  • The visual-search model. The packaged beta builds ship the image-understanding model inside the app, so nothing needs downloading. For developers running from source, the model is fetched only when explicitly requested (the launcher's --vision on option); the default behavior never downloads it — visual search simply reports itself as unavailable until the model is present.

In both cases the download brings software or a model to your machine; your files and catalogue are not sent anywhere.