Overview
KOGENT is the new security model for agentic systems. It sits between your agent runtime and external systems, enforcing identity, policy, and cryptographic proof on every action.
What it is
Autonomous agents are useful. They already schedule meetings, handle payments, process documents, book flights, draft emails, and act on your behalf. With where AI is today, agents are capable enough to manage your finances, negotiate contracts, and make decisions you currently make yourself. Most of them have no accountability, no audit trail, and no way to prove what happened.
KOGENT changes that. It wraps every agent action in three things: a signed identity, a policy check, and a cryptographic receipt. Nothing runs without authorization. Nothing leaves your machine unverified.
What it ships with
KOGENT ships with two things: a security infrastructure stack and production agents built on top of it. The security stack handles identity, policy enforcement, data privacy, tamper-evident receipts, and private payments. It is available as a Python SDK and runs on any agent framework. The production agents are real daily tools — each one runs entirely through the security stack and demonstrates what that looks like in practice. Vigil, the calendar and email agent, is the first. More agents follow.
System positioning
KOGENT sits at a specific point in the agent stack:
User → Agent Runtime → KOGENT → External Systems
The agent runtime handles reasoning and task planning. KOGENT handles the control boundary: verifying identity before execution, enforcing policy before actions reach external systems, filtering data before it reaches any model, and sealing every outcome with a signed receipt.
The OpenClaw skill system
KOGENT is built around modular skills. There are two categories: the security infrastructure that every agent runs on, and the production agents that ship with KOGENT.
Security infrastructure
- Identity: cryptographic agent identity anchored on Ethereum via ERC-8004
- Verification: tamper-evident receipts for every action taken
- Rules: policy enforcement before every agent action
- Sieve: PII redaction before data touches any model
- Specter: private payments using zero-knowledge proofs
- Reputation: signed event ledger for tracking agent trust over time
Production agents
- Vigil: Your autonomous personal assistant. Integrates with Microsoft and Google calendar, email, and chat. Monitors continuously, surfaces what matters, and takes autonomous decisions you configure: setting alarms, clearing them when plans change, surfacing urgent emails, and briefing you before every meeting. Available for macOS.
Design principles
- Local-first. Run KOGENT fully local on your own machine. Nothing leaves your device. In Vigil's local mode, your calendar and email data never touches any external service.
- Verifiable by default. Every action produces a signed, auditable receipt.
- Open standard. Identity is built on ERC-8004: inspectable, extensible, not vendor-locked.