Know which AI agents exist, who owns them, and what they are allowed to do.

AgentBoundary keeps each agent's purpose, owner, runtime, service identity, tool access, data limits, approval rules, review date, and stop plan together.

Connector access is not represented as generally available. Private pilots begin with inventory, policy review, approval records, and dry-run stop planning.

Ask about an AgentBoundary pilot Try the governance drill

AI-agent sourceAgentBoundary

Agent identityPurpose, owner, runtime, service account

Permission boundaryTools, data, environments, approvals

Stop readinessReview date, logging, dry-run stop plan

Evidence recordReviewed agent passport

Governance workflow

Move AI agents from an informal list to accountable identities.

AgentBoundary is built for inventory, policy enforcement, approvals, review campaigns, kill-switch planning, and retained evidence.

1Agent passport

2Policy decision

3Human approval

4Kill-switch dry-run

5Audit evidence

What your team can govern

Know what each agent can do, who owns it, and how to stop it.

AgentBoundary centers governance on a reviewed agent record and clear policy decisions instead of informal spreadsheets or scattered prompt notes.

Agent passports

Track identity, owner, purpose, runtime, service account, tool scope, data boundaries, review date, and logging status.

Clear policy decisions

Record allow, require approval, deny, or alert-only decisions with the matching policy and next step.

Approval workflows

Route sensitive tool access, production changes, customer data access, and external-send capability through assigned roles.

Kill-switch planning

Dry-run disablement for service principals, OAuth grants, tool servers, CI/CD references, webhooks, and app access.

Boundary model

Govern tools, data, environments, and credentials by reference.

AgentBoundary keeps agent governance focused on accountable metadata, permission boundaries, approval state, and audit records. It stores credential references and safe fingerprints, not raw secrets or sensitive prompts.

Identity and ownership Every agent has an owner, purpose, runtime, business unit, review date, and service identity context.

Tool and data scopes Allowed tools, denied tools, sensitive capabilities, data boundaries, and environment boundaries stay visible.

Review and approval High-risk agents can require SecurityAdmin, AI governance, app owner, or auditor involvement before action.

KairnexEvidence export Export sanitized governance packages for evidence review without exposing raw secrets or full sensitive prompts.

Security boundaries

Defensive governance software only.

AgentBoundary is for inventory, review, policy enforcement, approvals, dry-run containment planning, auditability, and evidence. Live destructive actions require explicit approval and typed confirmation.

No raw API keys, OAuth refresh tokens, passwords, cookies, private keys, or hidden credentials stored

No full sensitive prompts stored or exported

Policy decisions are deterministic and auditable

Kill-switch execution is dry-run first and approval-gated

Tenant-scoped records, RBAC roles, audit logs, and review campaigns support security review