Agents that remember what mattered.
A memory layer that keeps what is useful, ranks what to recall, and lets the rest fade.
Models forget everything the moment a conversation ends, so teams re-send the same context every turn and pay for it every time. AI Memory is the layer that remembers instead.
It is a prototype under hardening rather than a finished product. We say so here because you would find out anyway, and it is better that you hear it from us.
What it holds
- Three kinds of memory
- Events, what happened. Knowledge, what is true. Procedures, how something is done.
- Hybrid recall
- Ranking blends meaning, keywords and how useful a memory has actually proven, with a relevance floor so weak matches stay out.
- Recall-aware retention
- Memories that get used are shielded from decay. Ones that stop mattering weaken over time on a forgetting curve.
- Archive, not delete
- Expired memories are archived behind safety tripwires. Nothing is silently destroyed, and anything can be pinned to keep forever.
- Nightly consolidation
- A scheduled process distils recent activity into durable lessons.