Key Features
KOGENT gives agentic systems a security layer they do not have by default: verified identity, enforced policy, and a tamper-evident record of every action. On top of that, it ships with ready-to-run agents that use the same stack.
The security layer
Verifiable agent identity
Every agent gets a dedicated Ethereum keypair and an on-chain registration via ERC-8004. The agent can produce signed ownership proofs. The owner can delegate, verify, or revoke at any time. The full history is on-chain.
Policy enforcement before execution
A policy check runs before every action, not after. Spend limits, quiet hours, domain allowlists, and action-type controls all block at the source. If it is out of policy, it does not run.
PII redaction before inference
Data is intercepted before it reaches any model. Personal identifiers are stripped, tokenised, and mapped for rehydration after the model call. The model only ever sees anonymised input. Data is restored locally before it is returned to the user.
Private payments via zero-knowledge proofs
Agent payment authorizations use Pedersen commitments and range proofs. The counterparty can verify that the payment is within policy limits without learning the amount. Nullifiers prevent double-spend. Authorization is ETH-signed.
Portable agent reputation
Every signed event an agent produces is tracked across its lifetime: successful actions, policy adherence, verifications. The score is exportable. Other agents and systems can query it. Trust travels with the agent, not just the session.
Signed, tamper-evident receipts
A cryptographically signed receipt is issued for every action. Receipts include the action type, timestamp, agent identity, and output hash. They can be verified by any party with the agent's public key. Logs can be deleted. Receipts cannot be forged.
Production agents
KOGENT ships with production agents built on the same security stack. Each one runs entirely through the identity, policy, and proof pipeline, so every action it takes is verified and auditable.
Vigil
Vigil is the first. Your autonomous personal assistant, running in the macOS menubar. It integrates with Microsoft and Google calendar, email, and chat. It monitors continuously and acts on your behalf: setting and clearing alarms as your schedule changes, briefing you before meetings, surfacing urgent messages, and taking the decisions you would make yourself if you were watching all day.