Research

Research notes from building CPU-first video intelligence.

Ayneye is a product surface for World-State, semantic refresh, evidence-grounded answers, CPU-first detection, and video-aware agents.

Technical notes

Early research themes that explain why Ayneye is different.

Why World-State beats raw video context

Repeated questions should reuse state.

Semantic Refresh for low-cost video intelligence

Refresh detector work only when the scene materially changes.

Evidence-grounded Ask Video

Answers should cite artifacts and evidence bundles.

CPU-first neural detection

Prove cost control before GPU expansion.

Video agents need state

Agents query tools and evidence, not screenshots.