Model supply chain
How models are evaluated and changed in the environment where they will be used.
Detector testing, promotion and rollback
A detector should enter service with evidence of its usefulness and a recorded decision. Shadow evaluation gives the team a way to examine results before they affect the production queue.
Stages
Local learning.
Cyberdroid AI Detection, available now to select organizations, learns baselines from telemetry on the organization’s own hardware.
Shadow testing.
Each of 21 detectors is disabled, in shadow or in production, and replay measures its precision, recall, stability and volume.
Rollback.
An AI Detection detector can return to shadow or be disabled for review. Output limits keep overflow available outside the main queue; promotion and rollback decisions remain recorded.
Local serving.
Cyberdroid AI Investigation, in early access, serves its language models on the organization’s own hardware.
Verify
Detector changes.
Every promotion and rollback is on record.
Routing metadata.
The routing record identifies the model used for a request. Any external model route is an explicit policy choice.
Limits
Page scope.
These summaries explain the product’s role. The proposed deployment identifies the enabled functions, supported configuration and operating responsibilities.
