The product in a paragraph

Every AI assistant you use today reprompts you about the same facts — your dietary preferences, the medications you take, the languages you speak, the addresses you live at, the pets you own. This is silly. The context you have already given one agent does not travel to the next. GeraMind is a personal context vault that stores this information once, encrypts it at rest, and exposes it to your agents through consent-scoped queries with full audit.

What is in the vault

Basic profile (name, languages, time zones, addresses)
Preferences (dietary, travel, accessibility, communication)
Health context (conditions, medications, allergies) — sensitive tier
Documents (passport, insurance, tax returns) — encrypted blobs
Interactions (past agent conversations, with TTL)
Relationships (family, colleagues, healthcare providers)
Subscriptions and accounts (OAuth tokens, not passwords)

What is not in the vault

Raw credentials. GeraMind never holds plaintext passwords.
Anything you do not put there. There is no magic harvest.
Training data. The vault is yours; its contents do not train our models.

How agents use it

Agents call GeraMind through a query layer that enforces three things: the scope of the query (which categories), the purpose (why), and the minimum necessary data shape. If your agent asks for "preferences relevant to booking a restaurant tonight", it gets "vegetarian, nut allergy, Italian preferred" — not your entire preference history.

How we plan to guarantee privacy

End-to-end encryption with user-held keys for sensitive tiers.
Per-query consent tokens — agents can’t cache broad scopes.
Full audit trail accessible to the user, not to operators.
Export-on-demand to a portable format (we are defining one).
Hard deletion with cryptographic attestation of removal.

How it fits the rest of the Gera stack

The obvious integrations: GeraClinic can retrieve relevant health context to pre-fill a consultation form; GeraEats can filter out allergens; GeraNexus-initiated transactions can consume vault data under explicit consent. None of these are automatic — the user decides, per category, whether each Gera product (or any third party) can read at all.

Timeline

Q3 2026 — public schema draft, reference SDK.
2027 — closed beta for Gera Prime subscribers.
2028 — interoperability with MCP memory servers and portable formats.
2030 — general availability with full portability guarantees.

How to follow along

Design posts go up on this blog. Schema drafts publish to /research. If you want first access to the beta, join the waitlist. We are especially keen to hear from people who have strong opinions about data sovereignty, agent safety, and cryptographic access control.