EXEMPLARExemplar is part of the NVIDIA Inception program.
NVIDIA InceptionExemplar is part of the Google for Startups program.
Google for StartupsGovernance & Compliance
Enterprise AI strategy has a governance job: make dangerous execution hard, and keep material actions attributable and interruptible. Exemplar puts policy in front of agent tool calls and operational change, enforced continuously across the full lifecycle.
What this solution covers
- Policy before execution — Allow, ask, or deny before bash, path, deploy, MCP, or data access — whether the actor is an IDE agent, a production agent, or a human runbook.
- Approvals and interruption — Require a human when the action is material. Sensitive targets, production writes, and irreversible commands do not run by default.
- Attributable audit trail — Who did what, with which agent, and whether it was allowed — structured evidence security and compliance can read.
Examples
Illustrative scenarios — agents and operational change.
Cursor or Claude Code proposes a destructive bash or an MCP write. Policy returns deny or ask before the command runs. The verdict is in the audit trail with the agent, the user, and the tool.
A database failover is requested at night. Policy requires approval by the on-call lead before execution, then records the decision and command output.
Security asks who ran what. Teams export evaluated actions — requestor, agent, allow/ask/deny, timestamps — from one system of record.
Governance is the company layer in front of Relay (IDE agents) and Marshal. Day 2 runbooks still sit behind the same policy — see self-service actions. For continuous evidence, see Agent Assurance.
Book a demo →At a glance
Policy
Checks before production actions execute
Approvals
Human gates for risky or sensitive operations
Evidence
Traceable execution records for audit and retros
Governed execution supports the three jobs of enterprise AI strategy: adopt useful AI quickly, make dangerous execution hard, and keep material actions attributable and interruptible.