One foundation.
A bigger future.

A public core for the models you choose.

The public edition is designed to work with standard models. It keeps memory separate from model choice, so useful knowledge can travel between supported agents instead of staying inside one conversation.

The foundation is retained knowledge, source-aware recall, and controlled agent access. V2 brings that story into focus: keep useful work, find its context, and carry it into the next run.

The V2 public release is being prepared. Its exact code, license, setup instructions, supported clients, and upgrade path will be published with the release.

Learning that becomes advanced memory models.

We are developing proprietary memory models, informed by approved learning from our own data. The goal is better decisions about what to remember and which context is useful for the next task.

That model development is separate from the public core. Proprietary training assets, model weights, and private histories remain private. Customer data-use terms will be agreed separately for any hosted service.

Advanced models are in development. This preview does not announce released weights, measured improvements, or a finished learning service.

MemroOS as a service.

A managed MemroOS service is planned for teams that want shared memory without operating the full stack themselves. The public edition and the commercial service will have clear, separate release scopes.

We are preparing a pilot. Availability, onboarding, pricing, support, and data-use terms will be confirmed before customers are invited.

Where things stand.

Release notes will distinguish what is available from what is planned. A publication date and tested upgrade instructions will be announced with the final build.

Read the V2 release brief or return to the overview.