AI agent governance: questions buyers ask
Last reviewed
Short answer: These are the questions buyers put to us most, answered directly and including where the answer is “not today”.
Can an AI agent deploy to production on its own?
By default, agents cannot execute production deploys. Deploys run on the platform’s servers and are unreachable from agent dispatch.
Can an agent approve its own work?
No one approves their own gate, whether agent or human. An agent can ask to deploy; it can never approve its own request.
How do I stop agents leaking API keys and secrets?
Agents never hold provider credentials; the platform performs the privileged action on their behalf. When a credential appears in an agent’s output, the stream is cut and a rotation gate opens. A credential or workspace file is refused before it can be committed.
What is the audit trail for agent actions?
Every blocked attempt is audited. Nerve keeps an admin and chat audit trail.
How do I know an agent’s “done” is true?
An agent cannot self-report a merge or approval that did not happen; the platform verifies completion claims before a task can close. The real build and full test suite must pass before the approval gate opens; a failing build never merges.
Is my data separated from other customers?
Tenant isolation is enforced at the database and fails closed: even the database owner role cannot read another tenant’s rows. Operational data is tenant-isolated and encrypted at rest, and inference can route to the customer’s own model endpoints.
Does Nerve find real vulnerabilities or sample data?
Nerve runs real vulnerability scanning, including dependency CVEs, SAST and secret detection, on every tenant, not mock findings. Dependency CVE data is sourced from the OSV.dev database.
Can I keep agent spend under control?
Nerve shows cost per agent and per model, with daily trends. Budgets warn and then pause for a human.
Can I use my own models?
Model choice is per agent and per project, including local models.
Does it work with our single sign-on?
Nerve supports OIDC single sign-on and SCIM 2.0 provisioning. Nerve does not offer SAML single sign-on today; OIDC and SCIM 2.0 are supported.
Can I run it in my own cluster or air-gapped?
Nerve is a managed service by default; on-premises is available for enterprise customers at significant cost. Nerve is not self-hosted by default, and it does not currently offer a fully air-gapped deployment.
Which clouds does it deploy to?
Nerve ships via AWS CodeBuild to EC2, with an any-cloud-ready seam. AWS is the one live production deployment target today; other cloud providers are not yet validated against live infrastructure.
Does it make us SOC 2 or HIPAA compliant?
Nerve produces compliance evidence mapped to the control families your auditor uses. Compliance evidence is not a certification: Nerve does not claim attestation against any named framework.
Will it automatically kill a compromised workload?
An eBPF sensor monitors the production host continuously, feeding runtime detections into the security dashboard. Runtime detection is continuous; enforcement is operator-armed.