Who Owns the Model of You? — 0set0set Skip to content A self model is a person-controlled, contextual collection of claims an AI<br>agent may consult, with consent, to personalize its behavior. It is not a full<br>identity, not an objective account of a person, and not a profile owned by a<br>platform. Alma is my experiment in owning<br>and governing that model outside any single product.

Agents are getting better at tools and memory, but the model they build of a<br>person usually stays trapped where it was created: inferred opaquely, weakly<br>portable, and hard to audit. I think that model should be inspectable,<br>correctable, portable, contextual, consented, and controlled by the person it<br>describes.

Alma approaches the problem with a small architectural core: claims, provenance,<br>grants, temporary readings, audit events, inspection and correction, and schema<br>versioning. The current Rust prototype is narrower and more concrete: event-log<br>Events, projected Facets, scope-based grants, consent-filtered readings,<br>always-on reconciliation with numeric confidence, a signed export bundle, and<br>conformance fixtures. The contribution is the composition, not a new<br>cryptographic primitive or identity standard.

The limits matter. Signatures prove origin and integrity, not truth. Access<br>control governs disclosure, not downstream retention. And there is no evidence<br>yet that Alma makes agents more useful. The evaluation in this essay is a plan<br>for testing that hypothesis, not a result.

Key takeaways

A self model is a person-controlled, contextual set of claims for agent personalization — not a platform-owned profile, and not a full identity.

Memory records what happened; a self model represents how a person works, with provenance, context, uncertainty, and consent.

Alma Core is small as an architecture; the current Rust prototype uses a narrower 2026-06 wire format built from events, facets, grants, readings, and signed bundles.

Access control governs disclosure: a signature proves origin and integrity, not truth, and nothing controls data after it is disclosed.

That a person-owned model improves agents is a hypothesis ; this essay proposes a concrete first experiment, not results.

1. Introduction#

Every conversation with an AI starts the same way: with me explaining myself<br>again.

Who I am. How I work. How I like answers — short when the question is simple,<br>detailed when the system is complex. What I am building, and why. I type it into<br>a new chat, paste it into a “custom instructions” box, or repeat it because the<br>last assistant that learned it lives inside another company’s product.

We have taught machines to reason. We still keep making them meet me as a<br>stranger.

This is more than an inconvenience. As agents gain autonomy — booking, buying,<br>drafting, deciding — the model they use to act for a person becomes<br>infrastructure. And here is the uncomfortable part: platforms will build that<br>model whether or not we design for it. They already infer preferences and<br>patterns. The question is not whether a person gets modeled. The question is<br>who controls the model, who can see it, and who can move or delete it.

The argument runs in seven steps. (1) Useful agents need some model<br>of the person they serve. (2) Platforms inevitably build such models. (3) Those<br>models are usually opaque, platform-bound, and hard to audit. (4) Simple memory —<br>transcripts, vectors, a free-text profile — does not fix this, because it does<br>not separate durable preferences from passing mood, verified facts from guesses,<br>or shareable context from private values. (5) A person-controlled alternative is<br>possible, in which the model is inspectable, correctable, portable, contextual,<br>and consented. (6) Alma is an experimental attempt at that alternative. (7) The<br>risks remain significant, and the central benefit is still a hypothesis that has<br>to be tested.

The thesis is normative and technical at once:

If an AI agent builds a model of a person, that model should be inspectable,<br>correctable, portable, contextual, consented, and controlled by the person it<br>describes.

This essay combines a reference architecture, an experimental implementation, a<br>proposed interoperable surface, and a research agenda. It is not a standard, a<br>validated result, or a privacy guarantee.

2. The Structural Problem#

The difficulty is not a missing feature. It is structural.

Agent memory is fragmented. What one assistant learns rarely transfers<br>cleanly to another.

Platform models are opaque. Systems infer a great deal, but the person<br>typically has limited inspection, correction, provenance, and export.

Custom instructions are too flat. They help, yet they collapse context,<br>scope, temporality, and evidence into one block of text.

Transcript and vector memory are undifferentiated. They can retrieve what<br>was said, but they do not, on their own, distinguish a standing preference from<br>a one-off request, or a verified fact from an agent’s guess.

Consent is underspecified. “Use memory” is...