Trust
Security Practices
Operational and product security practices for authentication, logs, agent execution, and billing.
Security practices
This page describes practical controls we use and recommend when running agents in production.
This is not a compliance report. If you need a security questionnaire answered, see SOC compatibility.
Authentication
- User authentication is handled via Supabase JWTs.
- Public API access (v1) uses API key auth.
What you should do
- Rotate API keys regularly
- Never put secrets in prompts
- Use separate keys/environments for staging vs prod
Execution and isolation
All agents on Corgent are first-party and built by us; there is no third-party agent upload path.
- Agents run with minimal permissions
- Assume outputs can be wrong (human-in-the-loop for high impact actions)
Manager approval modes (web)
Manual: write actions are blocked until an explicit user approval is provided.Auto allow: write actions can execute immediately if write permissions are enabled.- Keep write permissions disabled by default in production unless your team has tested rollback paths.
- Re-check approval mode after role or team changes.
AI notice for automated execution
- AI-generated plans can be incomplete or incorrect.
- Treat auto-executed write actions as operator-assisted, not fully autonomous.
- For high-risk workflows (billing, deletions, customer communication), keep Manual mode enabled.
Logging and PII
Corgent applies PII redaction principles to logs so operational visibility does not become a data leak.
Recommended logging pattern
- Request id
- Agent id + version
- High-level outcome (success/failure)
- Timing + token usage
Avoid logging:
- raw customer messages
- credentials
- payment details
Billing and payments
Billing is computed from usage returned by the model/provider and then translated into credits.
Recommended approach
- Set budget limits for automation paths
- Use caching for repeated prompts
Incident response
If you suspect misuse or a leak:
- Rotate keys
- Disable affected agents
- Audit logs for scope
- Contact support with timestamps and request ids
Need more detail?
Start in the product, or write to us.