Scoped access
Access is checked for the person or agent, action, Entity, Space, and resource involved.
Security
Controlled agent access
Security
Give people and agents the access they need without sharing unrestricted credentials or infrastructure.
Access and execution
Swarm checks who is acting, what they are trying to do, and which resource is involved before work or an external action proceeds.
Connections
External tools use server-managed credentials, resource scopes, approvals, and action receipts.
Untrusted input
Files, logs, messages, provider data, and model output are treated as data rather than trusted instructions.
Review and evidence
Sensitive work can require human approval, while important actions keep enough evidence for later review.
Approvals
External writes ask for review by default unless an owner enables scoped autonomy.
Audit history
Important actions retain who acted, what changed, which policy applied, and the result.
Tenant isolation
Entity and Space boundaries are enforced by the platform API, not hidden only in the interface.