Sovereign memory
An agent's memory is encrypted on-chain under keys only the user holds — and deletion is a one-way door you can prove, not a promise you're told happened.
An autonomous agent is only useful if it remembers — your context, your preferences, what it did last week. But memory is also the attack surface: an agent that leaks its memory can be hijacked. Omyra makes memory sovereign — it belongs to the user, cryptographically.
Encrypted under your keys
Each memory entry is encrypted under keys derived from your wallet and committed on-chain. The operator stores ciphertext and a commitment; it never holds the key. Commitments use ZK-friendly primitives so they can be proven without being revealed:
- Poseidon hashing — a hash designed to be cheap inside a SNARK circuit.
- Pedersen commitments — bind a value without revealing it.
- Sparse Merkle trees — efficient inclusion proofs over a large memory set.
Provable deletion
Deletion spends a one-time nullifier into an on-chain account — a one-way door. Once spent, the entry is cryptographically unrecoverable, and a zero-knowledge proof attests the deletion without revealing what was deleted.
That is the difference between "we deleted your data" and a deletion you can verify. It is the GDPR right-to-erasure answer made into math — see the Vault Console for the product, and the Vault SDK for the API.
For an enterprise, "where did this data go, and can you prove it's gone?" becomes answerable. For an individual, the custodian who could read your memory simply does not exist.