A portable memory layer for human–AI continuity

Problem

AI assistants are increasingly useful, but their memory is fragmented by account, vendor, conversation, and device. People repeatedly restate preferences, projects, decisions, and boundaries. Centralizing everything in another opaque database solves portability by creating a new custody problem.

Design thesis

A useful v1 needs only three layers: a human-readable summary under the user's control, a deterministic one-way digest, and a wallet-linked receipt. The raw content and the proof have different privacy properties and must never be confused.

Data boundary

The current digest is SHA-256 over a canonical title, newline, and summary. The interface computes it with the browser's Web Crypto API before a transaction is constructed.

Proof state machine

1. Local: a digest exists only on the device.
2. Pending: the wallet returned a transaction identifier, but confirmation has not arrived.
3. Confirmed: a receipt reports successful execution.
4. Failed: a rejection, timeout, or reverted receipt prevents success.

Profile and points

A connected address is only eligible; activation is a separate user action. Standard activation creates a profile with 100 points. An optional, separately confirmed upgrade sends exactly 0.01 native coin and brings the profile to 200 points. Points are non-transferable, have no cash value, and do not promise a token or airdrop.

Registry contract

The minimal registry records profile activation, proof hashes, points, and the optional upgrade. Duplicate activation and duplicate boosts revert. The contract does not accept arbitrary raw memory strings. A withdrawal path is owner-controlled and does not affect user proof data.

Threat model

• A hash proves that some canonical content produced it; it does not prove the statement is true.
• Low-entropy content may be guessable, so summaries should not contain secrets.
• A compromised browser or wallet can expose local content or authorize unwanted transactions.
• Public transaction metadata is permanent and linkable to an address.

Scope and roadmap

Version one intentionally excludes automatic assistant scraping, background cross-platform synchronization, production vector storage, autonomous agents, and token trading. The next useful steps are user-controlled export formats, encrypted opt-in backup, permissioned sharing, and additional wallet-compatible networks.