Skip to main content
Security

Security first. Always.

AI agents need access to work, but that access must be controlled. We design every agent with least-privilege principles, human approvals, and comprehensive audit logging.

Principles

How we keep your data safe

Least-privilege access

Each agent only has access to the specific data and systems it needs for its task. No broad permissions, no "just in case" access. We map and document every data flow.

Read-only by default

Agents start with read-only access. Write access is granted only when needed for specific tasks, with explicit guardrails on what can be modified.

Credentials in a vault

All passwords, API keys, and tokens are stored in an encrypted vault. They are never included in prompts, never logged, and never visible to anyone.

You approve every send

By default, agents draft outbound communications and wait for your approval before sending. You control what goes out. Auto-send is opt-in for specific low-risk tasks.

Comprehensive audit logs

Every action is logged with timestamps, inputs, and outputs. Logs are reviewed monthly. You can request access to your logs anytime.

Your data stays yours

We map personal information flows and apply data minimization. Data stays in your accounts. Retention policies are built in.

Guardrails

What an agent cannot do

Every agent has hard limits on what it can access and do. These guardrails are set during the Build phase and cannot be overridden by the agent itself.

  • Cannot access systems or data outside its defined scope
  • Cannot send external communications without approval (unless explicitly unlocked)
  • Cannot modify credentials or security settings
  • Cannot delete data without explicit authorization
  • Cannot access other clients' data or systems
  • Cannot bypass logging or monitoring

Before go-live checklist

  • Data flows documented and approved
  • Permissions follow least-privilege principle
  • Credentials stored in encrypted vault
  • Approval workflows configured and tested
  • Audit logging enabled and verified
  • Guardrails tested with edge cases
  • Privacy impact assessment complete
  • Incident response plan documented
  • Team trained on approvals and escalation
FAQ

Security questions

Can an agent access data it does not need?

No. Each agent is configured with least-privilege access to only the specific data required for its tasks. We map every data flow during the Audit and document exactly what each agent can see and do.

Can an agent send emails or make calls without approval?

By default, no. Agents draft messages and wait for human approval before sending. You can choose to unlock auto-send for specific low-risk, high-volume tasks once you trust the output. Every send is logged.

Where are credentials stored?

All credentials, API keys, and login information are stored in an encrypted secrets vault. They are never passed in prompts, never logged, and never accessible to end users. Access is audit-logged.

How do you protect personal information?

We map personal information flows during the Audit. Agents only access, process, and store personal data for defined purposes. We apply data minimization and document what data is used and why.

Can I see what the agents are doing?

Yes. Every action is logged with timestamps, inputs, and outputs. You have access to audit logs on request, and we review them monthly as part of the Run service. Anomalies trigger alerts.

What happens if something goes wrong?

Agents have guardrails that prevent actions outside their defined scope. If an anomaly is detected, the agent pauses and alerts our team. We investigate, remediate, and report to you.

Questions about security?

We are happy to walk through our security practices in detail. Book a call or request our security documentation.

Book a call Read our privacy policy