How it works

From video source to evidence-backed answer.

The pipeline is intentionally boring: register a source, process bounded windows, create artifacts, expose evidence, and answer from state. Boring is good because buyers can audit it, developers can integrate it, and operators can reason about limits.

Overview How it works World-State Evidence Ask Video Accuracy Readiness Transparency Dashboard Artifacts

End-to-end product pipeline

Step	Input	Output	Buyer question
1. Add source	URL, HLS stream, uploaded metadata, or demo workspace	video record and tenant ownership	Can we control what enters the workspace?
2. Capture bounded windows	Configured duration, sampling, CPU-first detector mode	capture_report and detector_report	Can we cap cost and expose uncertainty?
3. Materialize World-State	Frames, events, zones, object observations	scene_graph.json and timeline.json	Can software query this without rewatching video?
4. Build evidence	Time spans, evidence refs, confidence, review notes	evidence_bundle.json	Can every answer show what supported it?
5. Ask from state	Question, artifact IDs, evidence-required flag	answer trace, citations, review_required	Can the system avoid guessing?
6. Inspect cost	Video duration, mode, fallback, retention	cost.json	Can finance understand unit economics?

What happens when scenes are hard

Hard footage does not disappear in the interface. Low light, crowded scenes, camera motion, low confidence, or ambiguous events should become visible review states. The product should not turn uncertainty into confident prose. It should show detector flags, fallback posture, evidence coverage, and a clear instruction that the workflow needs review or escalation.

Runtime

Bounded processing and detector reports are the operational core.

Open ->Quality

Accuracy

Accuracy is measured by artifact quality and evidence coverage, not marketing adjectives.

Open ->Beta

Readiness

Readiness means signup, dashboard, artifacts, limits, and review paths are all visible.

Open ->

Implementation sequence for teams

Create a beta account and open the dashboard.
Create the demo workspace before adding customer footage.
Inspect the generated artifact bundle.
Run one representative source with bounded duration.
Ask the same operational question multiple ways.
Compare answers against evidence_refs and timeline spans.
Decide whether the use case is safe for a larger pilot.

Product boundaries are part of the product, not footnotes. Ayneye is not presented as a replacement for every GPU video foundation model, not a free-infinite-query engine, and not an autonomous surveillance decision system. The current beta path is controlled signup, dashboard, REST API, bounded processing, evidence artifacts, visible limits, and read-only agent/MCP-style evaluation. Hard scenes, identity-sensitive workflows, emergency response, physical access, discipline, and destructive actions require human review or remain blocked.