Live video at scale,
the moment it happens.
Every frame reaches every screen in milliseconds, on relays you run anywhere.
Main stage, night two
Cost this month
Relays
Nearglass is an open source framework for live video with the speed of a call and the scale of a broadcast. Stream from OBS, any camera or a robot, and every frame reaches browsers, phones, apps and AI models in milliseconds, on relays you run anywhere.
Live in three steps.
Everything a live product needs.
One SDK for live video, robots and AI
Stream to an audience, drive a robot, or feed a model, on the same relays.
Main stage, night two
Live for 1 h 42 minst_8f2c41d09a7b- Encoder
- Connected over SRT
- Video
- 1920x1080, 60 fps
- Bitrate in
- 5.8 Mbit/s
- Keyframes
- every 1.0 s
Excavator 3, quarry face
Live, you are drivingst_31a90c5e72d4- Driving
- You
- Watching
- agent-7, 2 people
- If the link drops
- Stop, hold position
- Recording
- On, training data
agent-7 loaded the truck and asked for a person at the rock face. You took the sticks; it watches and takes back when you release.
Match feed, player tracking
Live, model attachedst_c0d71b8e93fa- Subscriber
- player-tracker
- Reads
- newest frame only
- Writes
- results data track
- Runs on
- your GPU
for frame in sub: out.send(frame, model(frame.image))
Frames, not segments
Each frame ships the moment it is encoded. LL-HLS waits to fill a segment first.
Illustration, not to scale.
Any encoder in, every screen out
SRT, RTMP and WHIP from OBS, ffmpeg or hardware. Browsers, phones and apps out.
A player for every screen
A web element, a React component, or a C library for native apps.
<!-- any page --> <nearglass-player stream="st_8f2c41" publishable-key="pk_..." auto-join>
import { NearglassPlayer } from '@nearglass/react'; <NearglassPlayer stream={id} publishableKey="pk_..." autoJoin />
nearglass_conn c = { .relay_url = url, .token = tok };
nearglass_sub *s = nearglass_sub_new(&c, ns, "video", err, sizeof err);
nearglass_sub_set_frame_callback(s, on_frame, dec);Tokens and admission
Public, signed-in or invite only, with short-lived viewer tokens and caps per stream.
<!-- anyone on your page takes a seat --> <nearglass-player stream="st_8f2c41" publishable-key="pk_..." auto-join>
// your backend, after the viewer signs in POST /v1/streams/st_8f2c41/viewer-tokens { "viewer": "u_2291" } // a seat for 5 min
// mint only for people on your invite list if (invites.has(user)) token = await mint(user) else deny() // no token, no seat
Python for AI
The newest frame as a numpy array, and results tied to the exact frame.
import nearglass as ng sub = ng.Subscriber(RELAY, "live/st_1", token=T) for frame in sub: boxes = model(frame.image)Early access
Data that rides on the frame
Send sensor readings, controls or model results with the video, each tagged to the exact frame it belongs to.
Holds up on Wi-Fi and 4G
The stream stays live on a bad link. Stale frames are skipped, not buffered, so viewers stay at the live edge, and quality steps down instead of stalling.
Run it yourself.
Relays wherever your viewers are
Our cloud regions, your own racks in a colo, or inside one venue. Every viewer joins the nearest relay.
Control plane and console
Streams, viewers, spend and relays in one place, with an API under every screen.
We build the encoder pipeline for your boards
Our low-latency encoding on single-board computers, for robots, cameras and devices.
Available to design partnersWe deploy the whole stack on site
Relays, control plane and console in your facility. We ship it prebuilt, set it up, and can run it for you.
Available to design partnersFree under AGPL-3.0
Run it on your own hardware, with a commercial license when AGPL does not fit. Source coming soon; design partners get access now.
Or let us run it.
Nearglass Cloud, early access for design partners
Built to reach millions of viewers
Relays stack into a tree. Every frame leaves the origin for every edge at once.
Real-time at broadcast pricesPriced per viewer-hour
Pay for watching, not for seats in a call.
Per viewer-hour at 4 Mbit/s 1080p. Public list prices, read Oct 28, 2026.
Spend caps and usage
A monthly limit per project, alerts on the way, and a hard stop at the cap.
Global relays, no servers to run
We run the relays in many regions. Your viewers join the nearest one.
Support from the people who built it
The engineers who wrote the relays and the player help you ship.
Open source, forever.
AGPL-3.0, with a commercial license when you need one.
Build it with us.
Early access, direct engineering time, and a say in what ships next.
Investors: contact the founder