OvenMediaEngine: one port, two protocols, and a comment saying so
OvenMediaEngine (OME) is a Sub-Second Latency Live Streaming Server with Large-Scale and High-Definition. #WebRTC #LLHLS
At a glance
- What is it?
- OvenMediaEngine is a low-latency live streaming server that ingests over five protocols, transcodes to adaptive bitrate, and serves viewers over low-latency HLS and WebRTC with an embedded relay. It is licensed under AGPL 3.0 with an in-bound contributor agreement and an Enterprise edition listed beside it, its shipped configuration gives two protocols the same port, and one codec needs hardware.
- Who is it for?
- OvenMediaEngine fits a team that needs sub-second latency to a browser without running a general-purpose streaming stack, and the ingest and egress coverage is unusually wide for a single binary. Verify four things before you deploy it.
- Can I use it commercially?
- Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly C++, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- OvenMediaLabs/OvenMediaEngine3,305 ★ · +29
| Repository | 2026-09-24 | 2026-11-15 |
|---|---|---|
| OvenMediaLabs/OvenMediaEngine | 3.3k | 3.3k |
29 Stars in the last 52 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
AGPL, an in-bound agreement, and an Enterprise edition next door
The front page never states a licence. The repository is recorded as licensed under the strongest of the common copyleft variants, which is the one with a network clause: modify it, serve it to other people over a network, and publish your source. For a media server that clause is the entire commercial question, and it is not answered anywhere on the page. The tree carries the licence file plus a separate third-party licence file and a contributor agreement document, so an in-bound CLA is in force on top of it. The company's own site, listed in the about section, describes an Enterprise edition available on a marketplace. So the arrangement is a network-copyleft project with a contributor agreement and a paid edition beside it, and none of the three is mentioned in the readme.
The shipped configuration gives two protocols the same port
The compose file assigns the low-latency HLS stream port and the WebRTC signalling port the same number, and carries the identical comment on both lines saying the two should be different if both are enabled. So the default configuration is knowingly invalid for the case the file itself identifies, and it works only because one of the two is off by default. That is the kind of thing that surfaces the day someone turns on the feature it warns about. Two smaller things in the same block. There is a third port for the TLS variant of the same signalling service, marked optional unless TLS is on, and it appears in the compose file but not in the quickstart command. And the compose file opens with a version key that current tooling ignores.
The edge service exists because of port arithmetic
The compose file defines two services from one image and one build context, and the only structural difference between them is which host ports they bind. The origin takes the signalling and relay ports in the low three thousands and the four UDP candidate ports from ten thousand upward. The edge takes the same container ports shifted onto different host ports, four thousand three hundred and so on, so both can run on one machine. That is the origin and edge clustering structure reduced to its mechanics: two instances, disjoint host ports, a shared configuration format. The edge service also lacks the restart policy the origin has. The feature list calls the arrangement clustering and the compose file is the whole of it.
The host address is a manual step, and both services ship one
Every documented start path sets an environment variable for the machine's own address, and the quickstart shows it as a literal placeholder of the shape your host address goes:
docker run --name ome -d -e OME_HOST_IP=Your.HOST.IP.Address \
-p 1935:1935 -p 9999:9999/udp -p 9000:9000 -p 3333:3333 -p 3478:3478 -p 10000-10003:10000-10003/udp -p 10000:10000/tcp \
ovenmedialabs/ovenmediaengine:latestThat is not cosmetic. WebRTC clients need a candidate address the server advertises, and a wrong value produces a connection that fails in ways that read as a network problem rather than a configuration one. Note also the port count: one line, six protocols, and the four-port UDP range that exists for thread distribution rather than for a service. The compose file ships a concrete private address for the origin and a second, much shorter private address for the edge. Whether those are two machines or a truncated value is not resolvable from the file. Either way both are committed example values that someone will copy, and the edge one differs from the origin one by exactly the characters a truncated address would lose.
The configuration instructions point inside Docker's private storage
The second quickstart command mounts two named volumes for the origin and edge configuration, which is the right shape. Then the text tells you the files now live under a path inside the Docker daemon's own volume storage, names the two directories it expects to find there, and, if you want them elsewhere, tells you to create symbolic links pointing into that path. The prefix that marks it as internal to the daemon rather than as an interface, so the documented way to reach your own configuration files goes through it. A bind mount of a host directory into the configuration path, which the compose file shows commented out, avoids the whole problem and is the obvious thing the instructions skip.
One codec needs hardware, and the SRT section describes a container
The embedded transcoder lists five video outputs. Four are software encoders. One is marked hardware only, which is the only qualification anywhere in the feature list and it is a real constraint: a deployment without GPU acceleration cannot produce that format at all, and the README says nothing about which hardware. Audio has three outputs and no such note. Elsewhere the SRT section is worth reading for what it is not. Its sub-bullets list secure reliable transport, an MPEG-2 transport stream container, and audio and video muxed, which are the same three descriptions the legacy HLS section gives. Nothing in that section describes a latency mechanism, which is the entire reason anyone would reach for that protocol.
Admission goes to a webhook, playback to a signed policy
Access control is two mechanisms and the README expands neither. Admission webhooks mean the server asks an external endpoint whether to let a viewer in, which puts a network dependency and a latency cost in front of every session. Signed policy means playback is authorised by something the server issued rather than by a token the client presents. Both are real designs and both are one line each. The same brevity applies elsewhere. Three digital rights systems are named with nothing about keys, certificates or licensing. Monitoring is one word with no metric named. Thumbnail is one word. And the opening claim is that the server streams to hundreds of thousands of viewers, which appears in the repository description as well and has no reference deployment or measurement attached anywhere.
The version stamp needs a git checkout before the build
The Dockerfile documents three build arguments and three build modes, and the mode that builds from a local checkout carries a warning. The ignore file excludes the git directory, so a build from local source has no history, and the file instructs you to run a shell script first to keep the full version stamp in the binary. Build from the default source and the stamp comes from the network instead. That is two different version paths, one of which needs a manual pre-step, and the second one silently produces a binary that reports less than it should. Two other details from the same file and the tree. The base image is selected by conditional expansion on the GPU argument, choosing between a plain distribution and a vendor CUDA developer image. And the supported-platform list starts four long-term releases below the distribution the image is built from.
Editorial conclusion
OvenMediaEngine fits a team that needs sub-second latency to a browser without running a general-purpose streaming stack, and the ingest and egress coverage is unusually wide for a single binary. Verify four things before you deploy it. Read the licence, because it is the network copyleft and it is nowhere in the front page, and running a modified version behind a network has obligations attached. Decide whether the shared signalling port matters to you, because the shipped configuration assigns one port to two protocols and a comment says they must differ. Set the host address deliberately, because it is what the server advertises to browsers and a wrong value breaks connections in ways that look like a network fault. And check whether you have hardware acceleration, because one of the five video codecs cannot be encoded in software at all.
Frequently asked questions
What is OvenMediaEngine?
A streaming server for live media that ingests over WebRTC, SRT, RTMP, RTSP and MPEG transport streams, transcodes to adaptive bitrate with an embedded transcoder, and delivers to viewers over low-latency HLS and WebRTC. It includes an embedded relay and signalling server, clustering in an origin and edge shape, file recording, re-streaming, thumbnails and a REST interface.
How do I set up OvenMediaEngine?
With one container command that publishes seven ports and sets an environment variable for the machine's own address, or with the compose file that defines an origin and an edge service from the same image. The readme carries no other setup instructions; the quick start guide and the manual are both hosted outside the repository.
What license is OvenMediaEngine released under?
The network copyleft variant of the GNU Affero licence, with a separate third-party licence file beside it and a contributor agreement document in the repository root. The front page states no licence at all. The company's site also lists an Enterprise edition on a marketplace alongside the open-source project.
Does OvenMediaEngine need a GPU?
Not for most of its work. Of the five video formats the embedded transcoder lists, four are software encoders and one is marked hardware only, so that single format requires hardware acceleration and the readme names no specific device. There is a documented build argument that selects a vendor CUDA base image instead of the plain distribution image.
How does OvenMediaEngine authorise viewers?
Two mechanisms, each described in one line: admission webhooks, where the server asks an external endpoint before letting a viewer in, and a signed policy, where playback is authorised by something the server issued. The readme expands neither, and it names three digital rights systems without saying anything about keys or licensing.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/ovenmedialabs-ovenmediaengine)Star history chart for your README
[](https://hysenlabs.com/projects/ovenmedialabs-ovenmediaengine)Was this useful?