Unimatrix is shared memory for the AI tools you connect. A client that calls the memory tools can store a memory, and another connected client can recall it. Unimatrix does not capture every conversation automatically. Desktop, iOS, and Android installers are not published.
A managed service runs on one host we operate. Self-host materials are provided after an enterprise agreement. Memory content is encrypted at rest with AES-256-GCM. Public traffic uses HTTPS. The TLS version is negotiated by the client and the deployment. Memory access is explicit — clients call memory tools rather than silently scanning user data.
Unimatrix name, logos, and mark are trademarks. Please use assets without modification.
Captured from the live product site. Click to view full size. Right-click / long-press to save.
Home — Unimatrix landing pagehttps://www.deployunimatrix.com/
Quickstart — start using Unimatrixhttps://www.deployunimatrix.com/quickstart
MCP Reference — API documentationhttps://www.deployunimatrix.com/docs/mcp
Help page for the unpublished mobile client. It is not an App Store or Play Store product.https://www.deployunimatrix.com/help/mobile-appPublic listings for the hosted MCP server.
Press and interview requests: hello@deployunimatrix.com
Email hello@deployunimatrix.com. This page does not promise a response time.
Most AI memory today is siloed per product. Unimatrix is a shared substrate: the memory is not owned by any one AI client, so any client can read and write it through one standard interface (MCP).
Memory content is encrypted at rest with AES-256-GCM. Like most hosted services, the server decrypts content transiently during ingestion (to build search indexes); memories composed in the dashboard can be encrypted in the browser before upload. Processing is strictly confined to your tenant with isolated encryption keys, and we never train models on or sell user memories.
Yes. The full stack ships as Docker images with a self-hosted deploy path, so teams that need full data control can run Unimatrix on their own hardware.