The continuity layer between the trackers that hold your real-life state and whatever AI workspace you already use. A browser add-on and web surface, not another assistant. Its whole job is to keep those two things connected without letting either one quietly swallow the other.
A tracker is not useful if you cannot reach it while life is actually happening. You are holding a baby, or standing in a school car park, or halfway through a conversation with an assistant that already has the context and cannot write anything down. The data exists. It is simply somewhere else.
MindTether is the bridge. You can open it and update a tracker directly. Or you can say something ordinary in the AI platform you were already using, and it updates the right tracker through a connector you approved. Then it can pull that information back and compile it into something worth reading.
The product is not the AI. Plenty of things want to be your assistant. MindTether wants to be the controlled boundary between the assistant you chose and the records you own, so that the living state stays in one piece instead of drifting apart across four apps and a chat window.
MindTether is the first consumer of these trackers, written by the same person, running on the same infrastructure. That is exactly why the boundary gets written down: the temptation is to grant it everything on the grounds that it is us, and to discover a decade later that "aggregation layer" quietly meant a second copy of the most sensitive record in the house, on a machine with different backups and a different threat model.
It is a specific, listed set of permissions, and the answer to "what does this get" has to be a list somebody can read, not "whatever the API returns". If the honest answer is ever everything, that should be a decision taken once, deliberately, with the word in front of it, rather than the default that happens because splitting it was fiddly.
Redaction is not concealment. A consumer without an elevated scope gets the row with its value withheld and a flag saying so, rather than no row at all. A consumer that cannot tell "not permitted" from "did not happen" will draw exactly the wrong conclusion, and that failure is silent.
Counts are a smaller disclosure than contents. Findings carry totals and references rather than the entries themselves, and following a reference needs a scope of its own. So a consumer can be told that a pattern exists without being handed the log it was computed from.
MindTether is the reason every tracker in this collection has a documented HTTP API, a scoped token system and a working export before it has a marketing page. None of that was built speculatively. It was built because something was always going to need to read across all of them, and the alternative to designing that boundary deliberately is discovering it by accident.
Each tracker describes itself at /v1/openapi.json, generated from the
same schemas that validate its requests, so an integration cannot be written against
documentation that has drifted. Tokens are minted and revoked from the tracker's own
settings, by the person whose data it is, and every one of them carries an
enumerated subset of scopes rather than a blanket grant.
MindTether is an approved working name and a written product direction. There is no shipped code and no date. The permission model is explicitly a draft, published internally to be argued with, on the reasoning that a permission model nobody disagreed about is one nobody read.
It is on this page because it explains the others. If you want to argue with the boundary before it hardens, that is the most useful thing you could do with it: archon@lockewerks.com.