Govern what your AI agents are allowed to do — without touching their source code.
KRNLA sits between your agents and everything they talk to. Register your agents, define structured policies in the dashboard, and every request is authenticated, inspected, evaluated and enforced at the gateway before it reaches its destination.
Policies live in the control plane, not in your agents. Detection is deterministic and runs in-process — no second model in the hot path. Telemetry is emitted asynchronously and carries metadata only.
Agent registry
Register every agent, issue a scoped credential, rotate or revoke it. Secrets are hashed and shown once.
Structured policies
WHEN a condition holds, FOR a scope, THEN allow, block or redact. Machine-readable, never free text.
Agent-to-agent
KRNLA mediates delivery between your own agents, so a pair can be authorized centrally.
Streaming aware
LLM responses are governed chunk by chunk instead of being buffered in memory.
Metadata telemetry
Security events keep decisions, reasons and latency. Prompts, responses and PII values are not stored.
Tenant isolation
Organization identity is derived from credentials only — never from a field in the request.