Schemas
Artifact family for video World-State integrations.
Schemas explain how Ayneye turns footage into stable files your application can store, query, render, and audit.
Artifact family
| Artifact | Contains | Used for |
|---|---|---|
| scene_graph.json | Objects, events, zones, relationships, confidence, evidence refs | Application state and agent memory |
| timeline.json | Ordered events and semantic refresh markers | Timeline UI and replay |
| evidence_bundle.json | Frame/time spans, answer traces, review notes | Trust and human review |
| detector_report.json | Detector mode, thresholds, warnings, fallback | Quality and escalation |
| cost.json | Duration, usage, retention, fallback, estimate | Budget visibility |
scene_graph.json sketch
{
"video_id": "vid_...",
"objects": [{"id":"obj_1","label":"vehicle","confidence":0.82}],
"events": [{"type":"entered_zone","t_start":10.2,"evidence_refs":["ev_1"]}],
"review_flags": ["low_light"]
}Schema handling rules
- Treat artifact_version as immutable for a given execution.
- Store evidence_refs with any answer shown to users.
- Do not merge outputs from different executions without tracking provenance.
- Handle missing or low-confidence fields explicitly in the UI.
- Use detector_report warnings to decide whether human review or GPU escalation is needed.