Binge

Real-time video understanding for AI agents.

Open-source reference app showing what’s possible with Fabric (compute) and optr (runtime) for real-time streaming AI. An agent points at any live stream — watch, understand, search, react — all through a simple API, paid per use with x402.

What It Can Do

• Caption — continuous text describing what’s happening
• Index — searchable memory of the entire stream
• Extract — pull structured JSON from every frame
• Monitor — register conditions, get webhooks when they hit
• Detect — bounding boxes, keypoints, segmentation
• Generate — summaries, reports, social posts
• Clip — cut video segments from search results
• Dual-tier — run a cheap model continuously, trigger an expensive one only when something interesting happens

API

POST /sessions
{ "url": "https://youtube.com/live/abc", "pipeline": ["caption", "index", "monitor"] }

GET /sessions/ses_abc123/search?q=when+did+the+guest+arrive

POST /sessions/ses_abc123/prompt
{ "text": "Is the host bullish or bearish?" }

Supports YouTube, Twitch, Kick, RTMP, RTSP, HLS, SRT, WebRTC, pump.fun streams, and any URL with a video stream.

What It Proves

Built entirely on Fabric and optr. Fabric provisions GPUs and CPU workers. optr connects the processing nodes — capture, VLM, Whisper, embedding, storage — through low-latency QUIC streams. Proves the stack handles high-throughput, low-delay, pay-per-use AI services at scale.

Status: Product complete. Open-source. Awaiting market timing for active launch.