Archivist pulls your content out of cloud services and devices, such as Google Drive, mail, local folders, an iPhone over USB, Takeout exports, messages and bank exports, and organizes it into one library folder you choose, described by plain markdown any person or model can read without the app.
What it does
Four invariants hold the design together. Files on disk and an append-only journal are the only authoritative state; the catalog, thumbnails, embeddings and markdown are rebuildable caches, and a daily catalog backup protects what a rebuild cannot recover, such as review history. Originals are immutable and nothing is ever auto-deleted; only byte-identical duplicates are skipped and anything ambiguous goes to human review. Every change to the library passes one journaled, two-phase choke point. And the library stands alone: files plus markdown stay legible if the app disappears tomorrow.
Beyond storing files, Archivist contextualizes them. Transactions become queryable, receipts are matched to the bank line they explain for you to confirm, and you can group messaging handles under the people behind them. Analysis runs on your own machine, using a local model the app downloads once on first use.
The default posture is to read local bytes and hold no cloud credential at all. Connectors that keep a standing cloud credential are fully supported but sit behind a plain-language disclosure of the trade-off.
How it treats your data
Local-first by construction. No telemetry. Opt-in cloud AI, when used, is journaled and capped.
Where it stands
In development and not yet released. Builds exist for Apple silicon Macs, Windows and Linux, but no one outside the studio is using it yet. It is the desktop side of the Human Memory work, and the waitlist on human-memory.ai is for it.
Archivist is made by Ocean Dweller Media. Questions and bug reports: support@human-memory.ai.