In development

PRODUCT 02

Memory Layer

A hosted, multi-client memory service for AI systems. Privacy and non-mixing are the constraints the service is designed around.

Memory Layer is in development and is not available. There is no waitlist number, no beta, and no published date. This page describes intent, not a shipping product.

An assistant that forgets is a tool you reintroduce yourself to every morning. An assistant that remembers across clients, without mixing what belongs to whom, is a different kind of instrument. The hard part is not storage. The hard part is isolation.

Multi-client by design

One memory service, many AI clients. The point is continuity across the tools someone already uses rather than another silo that only one assistant can read.

Non-mixing as a constraint

Separation between contexts is the design constraint the service is built around rather than a setting applied afterwards. A memory service that leaks between contexts is worse than no memory service.

Privacy first, in the same sense as the product

The privacy position is the one already published on finalverdict.ai: content is not training data, and deletion means deletion. Memory raises the stakes on that position rather than changing it.

Memory Layer: continuity across clients, isolation between contextsSeveral AI clients read and write to one hosted memory service. Each context is separated, and there is no path between contexts.CLIENTSClient AClient BClient CMEMORY LAYEROne service. Many clients.Context 1Readable and writable by its own client onlyContext 2No path to context 1Context 3No path to context 1 or 2Non-mixing is a designconstraint, not a setting.The hard part is not storage. The hard part is isolation.
Isolation is the product. Continuity across clients, with no path between contexts.

Interested in Memory Layer?

We are talking to a small number of organisations about early use. No date is committed.

Contact