Research note
The rise of personal AI memory
The interesting application of machine understanding, for most people, is not a general assistant. It's a system that understands the specific, unglamorous material they've already accumulated.
The category is young enough that the distinctions still matter more than the branding.
01 · Distinctions
Four questions separate these products.
Almost every tool in this space can be placed by answering: where does the understanding run, what does it capture, does it produce connections or just answers, and who holds the data.
The answers imply very different products, even when the marketing language is nearly identical.
- Where it runs
- On device, in a private cloud, or in a general model's context.
- What it ingests
- Only what you deliberately save, or everything your screen sees.
- What it produces
- Answers to questions, or a navigable structure over your material.
- Who holds it
- You, a vendor, or a vendor plus a model provider.
02 · Scepticism
Two claims worth distrusting.
The first is total capture: systems that record everything you see. The recall demos are impressive and the failure mode is severe, because an archive you didn't curate contains things you'd never have chosen to keep.
The second is that a chat interface is the answer. Conversation is a fine way to ask a question and a poor way to browse, recognise or wander, which is most of what people actually do with their own material.
Ask where it runs, what it takes, and who ends up holding it.
03 · Position
Where Shelf sits, plainly.
Shelf captures only what you deliberately send it, does its understanding on your Mac, produces a navigable library with visible connections rather than a chat window, and keeps the library and its index on your machine with no account and no sync server.
That's a set of trade-offs, not a claim of superiority. It costs you cross-device sync and the convenience of the largest cloud models, and it's the right trade only if you value the archive staying yours.
Next thought
Underneath all of it is one missing piece of the operating system.
Read: the missing memory layerIf this idea interested you…
Shelf V3 is out now
Shelf V3 is the current build. Shelf does not use automatic updates, so download the latest build here and install it over your current version.
This is intentional. Shelf is built around a local-first, privacy-focused philosophy. An automatic updater requires additional background processes, network connections, update infrastructure, and another system with permission to modify the app. We prefer to keep Shelf’s network footprint and background activity as small as possible.
Manual updates keep the process explicit: Shelf only checks the internet when a feature actually needs it, and you decide when a new version is downloaded and installed.