Security
Crawbl is built for organizations that need strong security controls over AI agent behavior.
Trust Model
Three trust levels govern what the system can do on a user's behalf.
- Observe
- Ask before write
- Trusted automation
Read-only. Crawbl reads and summarizes but cannot change external systems.
Before creating or updating anything, Crawbl shows a clear proposal. This is the default.
Bounded actions can run automatically under explicit limits and policies.
Zero Trust Architecture
| Principle | Implementation |
|---|---|
| Never trust, always verify | All requests authenticated, all services verify each other |
| Least privilege | Components get minimum permissions needed |
| Explicit verification | No implicit trust based on network location |
| Assume breach | Segmentation limits blast radius |
Internal service communication is mutually authenticated and encrypted.
Secret Management
All secrets are stored externally -- never in Git.
| Secret Type | Storage | Access |
|---|---|---|
| API Keys | Managed secrets service | Auto-synced to runtime |
| OAuth Tokens | Managed secrets service | Per-user, encrypted |
| Database Credentials | Managed secrets service | Injected via K8s secrets |
| TLS Certificates | Automated cert management | Auto-renewed |
Secrets are automatically rotated where possible, scoped to specific workspaces, and audited for access.
Network Security
Agent runtimes have no direct internet access.
Internal service communication is mutually authenticated and encrypted. Network policies restrict traffic to authorized paths only.
This is enforced at the network level. Even a compromised agent pod cannot reach external services -- all outbound access goes through the orchestrator, which enforces permissions and logs every call.
Compliance
- Audit trail -- Every action logged with timestamp, user identity, action type, and result
- Data residency -- Deploys in your infrastructure; no data leaves your environment
- Certifications -- Designed to support SOC 2 Type II, HIPAA, GDPR, and ISO 27001
Deep Dive
- Authentication Flow - OAuth and auth implementation details