Traditional
Video → frames → vision model → tokens → repeated expensive queries.
AI-readable video intelligence
Turn video into AI-readable World-State.
Ayneye converts live or recorded video into compact scene state, evidence traces, and searchable answers — CPU-first, developer-ready, and designed to avoid sending raw frames to LLM workflows by default.
CPU-first runtimeEvidence-groundedONNX detector validatedPrivate beta previewNo raw frames to LLM by default
Video AI is expensive because most systems treat video as pixels.
Ayneye takes a different path: it turns video into evolving state, then lets applications, APIs, and agents query that state.
Ayneye
Video → World-State → Evidence → API/MCP Tools → governed agents.
A runtime for software-readable video.
Process video into durable artifacts once, then ask questions and build workflows against state and evidence.
Capture
Bounded CPU-first processing from streams or recordings.
Understand
Scene graph, semantic events, and TSR refresh decisions.
Ask
Grounded answers with traceable evidence artifacts.
Build
REST APIs today and MCP preview surfaces after P67 runtime.
Built on real runtime primitives, not a demo prompt.
Ayneye public pages reflect artifacts, evidence, CPU detector, tenancy, limits, and protected metrics.
Per-video artifacts
runtime_report.json, timeline.json, scene_graph.json, evidence_bundle.json, and cost.json.
Tenant boundaries
API keys, roles, usage limits, and beta controls are product primitives.
CPU neural detector
ONNX/OpenVINO-ready detector path with safe fallback posture.
Metrics protected
Operational metrics remain behind the staging boundary.
Designed for developers building video-aware agents.
Start with REST API preview, then connect MCP tools and agent workflows once P67 runtime is implemented and audited.
curl -X POST https://api.ayneye.com/api/videos/add \
-H "Authorization: Bearer $AYNEYE_API_KEY" \
-d '{"url":"https://example.com/live.m3u8"}'{
"video_id": "vid_...",
"world_state_ready": true,
"evidence_ready": true,
"ask_video_ready": true
}Use cases for AI-readable video.
Ayneye is built for developer workflows where video must become state, evidence, and governed tools.
Developer video agents
Let agents ask questions over video state instead of raw pixels.
Live stream intelligence
Turn streams into structured timeline and scene state without GPU-first architecture.
Security and event review
Find what happened and where the supporting evidence is.
Media operations
Convert long video into queryable scenes, timestamps, and summaries.
Research archives
Analyze recorded media as structured evidence instead of manual watching.
Build with AI-readable video.
Join the controlled private beta and help shape the developer API for World-State video intelligence.