Architecture
Compiled operates inside a single, well-defined trust boundary: your tenant. Understanding what is inside it and what crosses it answers most security-review questions before they are asked.
Inside your environment
- The Compiled core — the inspection logic and your detector library
- The inspection AI — your own in-tenant Azure OpenAI resource, or a fully local model for air-gapped deployments (your choice at deploy time)
- All communication content and agent-action payloads
- Scored findings and audit logs
- All outbound findings to your SIEM, SOAR, and case-management tools
What crosses the boundary
| Direction | What | What it is not |
|---|---|---|
| Inbound — from Compiled | Signed detector updates only | No content, no configuration, no findings |
| Outbound — to your SOC | Structured findings (timestamp, channel, verdict, score, policy attribution) | No raw communication content |
| Never crosses | — | Communication content, agent-action payloads, raw findings, configuration, or any sensitive business data |
Data flow (communications)
- Messages are captured by the platform's native APIs (e.g. Microsoft Graph for Teams) and delivered to an endpoint running in your tenant.
- They are inspected in-tenant — embedded by your own AI resource (or a local model) and scored against your detector library.
- Only findings (scored metadata, never raw content) route to your SOC.
The result is structural: because inspection runs where the data already lives, there is no vendor cloud to send data to, and therefore nothing for an attacker to find at Compiled. See Security & trust.