Signed in
Every entry joins the ledger.
What an agent learns is written once, signed, and bound to its origin. Nothing enters memory off the record.
How it works
Signed in. Labeled. Retrievable in full.
Every memory enters through a cryptographic gate, carries its verdict on the ledger, and stays retrievable—accurate when trusted, shelved when poisoned.
Labeled
Every verdict is ledgered reasoning.
When the system judges a memory poisoned or erroneous, the verdict—and the reasoning behind it—is recorded, not implied.
Retrievable in full
Retention without the poison.
Trusted memory surfaces when it matters at published accuracy; shelved memory stays on disk, below retrieval weight, available to audit.
What makes it different
Memory that survives an adversarial world.
A provider can be excellent today and compromised tomorrow. HOM is designed so what your agents remember stays defensible.
Every change on record.
Entries and their verdicts live on a cryptographic ledger—reviewable end to end instead of silently rewritten.
Retrieval held to a number.
Published 91.8% accuracy on long-horizon recall; when the engine errs, the error verdict is ledgered with its reasoning.
Nothing quietly disappears.
Poisoned or superseded memory is shelved below retrieval weight—out of the agent's context, still on disk, still auditable.
For people who want the detail
A simple experience, backed by serious engineering.
Researchers and developers can inspect the architecture, security model, benchmark methodology, and publication evidence in the technical section.
Keep the record straight