Glossary

Architecture

HarnexaHarness

HARNEXA's governed execution loop for every agent tool call.

Definition

HarnexaHarness is HARNEXA's control layer for agent execution. Every tool call moves through identity, permission, budget, approval, computational sensors, and append-only audit logging before the workflow is considered safe to operate.

Why it matters

Governance must be architectural. If controls live only in policy documents, a model or workflow can bypass them.

How HARNEXA uses it

HARNEXA uses the harness in PHANTOM, AgentOps proof, readiness passports, and client deployment architecture.

FAQ

Questions this definition answers.

Can the model bypass HarnexaHarness?

No. The harness is the execution path for tools; the model does not get direct uncontrolled access.

Which layers are inside the harness?

Identity, permissions, budget, approval gate, computational sensors, and append-only audit logging.

Related terms

GovernanceGoverned agentAn AI agent constrained by identity, permissions, risk classes, approval gates, sensors, and audit logs.GovernanceAppend-only audit logA database-backed event log where agent events are inserted but never updated or deleted.Risk classFINANCIAL risk classThe risk class for commercial, customer-impacting, or money-adjacent tool actions.

Need these definitions mapped to a live commerce workflow, agent estate, or protocol-readiness path?

Book diagnostic sprint