Join the waiting list
Early access

Point AI at live TV.
Ship what it builds.

Sceney is the interactive layer for live video. It watches and listens to any stream. Say what you want on top of it, and it is running on the stream in about 90 seconds: overlays, 2nd-screen games, shoppable moments, TV apps. People build by chat. AI agents build through MCP.

Get in

Live stream · example requests
Watching the streamWriting the codeChecking it on the streamLive
What it does

Video is the input. Software is the output.

01

It watches. It listens.

Local models follow every scene change and transcribe every word. Ask when something appeared or what was said, and it answers from what it already knows.

02

Say it. See it.

Describe an overlay, a panel or a whole experience in a sentence, or attach the deck that describes it. It is built, checked against the video, and repaired if it is wrong.

03

Point at the video.

Drag a box around a player, a product or a logo and say “this”. The request is about exactly what you pointed at.

04

Every phone is a controller.

Trivia, predictions, votes: viewers scan a code on the screen and play along. The game is generated from what is on air.

05

Shoppable by default.

It searches the Edge product catalog for what is in the frame, so what you build can put real products on the screen.

06

It knows the broadcast.

It reads ad breaks from SCTE-35 and the programme guide. What you build can change the moment a break starts and change back when play resumes.

Why it fits

A layer for the stack you already run.

Networks, platforms and device makers already own the stream, the screen and the audience. Sceney adds the part that turns them into software, and it asks for nothing to be rebuilt.

A

Nothing to re-encode.

It reads the HLS stream you already publish. There is no new encoder, no marker to insert and no change to the signal chain.

B

1 runtime on every screen.

Everything it builds runs on the same small web runtime. It exports as an Android app and as webOS and Tizen TV apps, and the TV remote operates it.

C

Perception at a flat cost.

Watching and listening run once per stream on models Edge hosts, and everything built on that stream shares the result. The 10th experience on a channel adds no perception cost.

D

Inside the player you already have.

1 script tag puts what you built on top of your own web player. Your player keeps the stream, the controls and the audience. Sceney adds the layer.

E

Measured from the first view.

Views, time on screen, taps, phones that joined and links opened, per experience and per screen, with the share that happened during ad breaks.

F

Tied to the moment on air.

SCTE-35 ad markers, the programme guide and Sceney's own frame fingerprint place every experience at an exact moment of the broadcast.

90 sfrom a sentence to software running on the stream
0changes to your stream or your encoder
3screens from 1 build: TV, web and phone
24/7hosted and operated by Edge once you approve it
How it works

You design it. We run it.

1

Point it at a stream

Paste the address of any HLS stream. It starts watching and listening at once.

2

Build by talking

Ask, point, attach a brief. Keep changing it until it is what you want. Export it as an Android or Android TV app to try it on a real screen.

3

Hand it to Edge

You see the monthly price to run it around the clock. Say the word and Edge hosts it, operates it and keeps it running.

For AI agents

Your agents can run all of it.

Everything a person does here by chat, an AI agent does through MCP, the open protocol that Claude Code, Grok and other agents use for tools. Give an agent a stream and a task. No screen needed.

MCP

It sees the stream.

An agent gets the frame on air, an account of the last seconds with sound, the transcript, the ad breaks and the programme guide.

MCP

It builds and ships.

It writes an overlay itself or asks Sceney to write it, puts it on channels, and exports apps for Android, webOS and Tizen.

MCP

It stays on watch.

A standing task runs on a stream every few minutes, at each ad break, or when a phrase is spoken, and keeps the results for the agent to read.

MCP

What you build is a tool too.

An overlay can offer tools of its own, at its own MCP address. Another company's bot starts the trivia round, shows the product, reads the score.

MCP

1 line to connect.

An address and an access token. Each token belongs to 1 workspace and is revoked by itself.

claude mcp add --transport http sceney 
MCP

Measured the same way.

The agent reads views, taps, watch time and the monthly price for what it built, and reports back to you.

Waiting list

No invite code yet?

Tell us who you are and what you would build. We are opening workspaces a few at a time. If you run a platform or build devices, say so, and we will show you how Sceney fits inside it.