Solid (solidproject.org)

Solid, Tim Berners-Lee’s project, defines a pod — a user-owned data store — accessed via WebID and Linked Data Notifications. It is the most principled and longest-running effort in this space, and its scope vocabulary (via RDF ontologies) is thoughtful.

Where it overlaps: the principle of user-owned, portable storage. Solid pods can hold structured personal data and expose it over a standard protocol.

Where it does not: Solid is storage-centric. It does not specify the purpose-bound query layer, the minimisation service, or the audit-log semantics that matter for agents asking small questions. Solid pods, in the base protocol, tend to return documents; GeraMind returns answers.

Our relationship: a Solid pod could be the storage backend under a GeraMind deployment. We are explicitly designing the query layer to be storage-agnostic for this reason.

Inrupt (inrupt.com)

Inrupt is the commercial company building enterprise-grade Solid pods and identity services. They have real customers (NHS, BBC pilots) and a battle-tested implementation.

Where it overlaps: enterprise personal-data custody.

Where it does not: the developer surface is still document-shaped. An enterprise integrating Inrupt pods is integrating a storage system, not a minimisation-first query service.

Our relationship: complementary. Inrupt hosts pods; GeraMind could be the query layer running in front of Inrupt pods for consenting customers.

Apple on-device context APIs

Apple has quietly built a rich surface — Health, Home, Passkeys, Wallet, AppIntents — that exposes narrow, scoped user data to apps on-device. These APIs are good: scope- limited, user-controlled, privacy-respecting.

Where it overlaps: on-device minimised answers to app queries.

Where it does not: device-bound, not portable across platforms or operators. If you change from iPhone to Android, or want your Mac agent and your web service to share context, Apple’s surface does not help.

Our relationship: complementary. An iOS client of GeraMind should read from Apple’s local APIs for the data Apple already exposes (where the user consents) and use the GeraMind vault for the rest.

Digi.me and Data Transfer Project

Digi.me (now defunct) and the Data Transfer Project focus on the portability problem — moving data between providers. They prove that structured cross-provider transfer is possible.

Where it overlaps: portability as a first-class concern.

Where it does not: the query / minimisation / audit layer is out of scope. Transfer projects answer “how do I move data from A to B?” GeraMind answers “how do I let an agent ask a scoped question across that data without moving it anywhere?”

Our relationship: orthogonal. DTP-style import adapters are how the GeraMind vault gets populated without manual re-entry.

ChatGPT Memory and MCP memory servers

We covered these in an earlier post. Short version: ChatGPT Memory is vendor-locked; MCP memory servers are per-agent silos with no shared scope vocabulary.

Where we are genuinely different

Three commitments these projects do not make in full:

Queries return answers, not documents.The minimisation engine is the default; raw retrieval is a second-class, higher-friction path.
Purpose is a first-class field. Every query has a purpose, every token binds to a purpose, every audit entry records a purpose. GDPR purpose-limitation becomes mechanically enforceable.
Audit is a user tool, not an ops tool.The log is designed for the person whose data it is, with readable entries, revoke actions, and weekly digests.

Where we might be wrong

Minimisation might not be enough. For some queries, an agent needs raw content; a minimisation-first design might force too much friction and push developers to request raw access by default.
Purpose binding is easy to game. A misleading purpose string is still technically valid. We rely on audit + reputation to make misuse expensive; this is not perfect.
Scope vocabulary is a coordination problem. We have an opinionated draft, but convergence with Solid ontologies, Apple’s taxonomies, and regulator categories is unfinished.

Cooperative future

The best outcome is layered: Solid / Inrupt for storage + identity, GeraMind for query + audit, Apple / Google on-device integrations for the data the OS already holds, DTP-style importers for population, and applications written against a single vault surface that abstracts these. We would rather be the right answer in the query layer than a mediocre answer across all of them.