An open source media server we have been building since 2009. This is what it does today, and what is coming, starting with an interface where you draw the pipeline instead of configuring it.
In the builds you can download right now.
MistServer reads the protocols your encoders and cameras send, and serves the ones your viewers and partners ask for. It repackages between them as the stream passes through, so there is no separate transmuxing step to run.
Insert timed ad-break cues into a live stream while it is running. A JSON request creates an SCTE-35 metadata track and MistServer carries the marker into its MPEG-TS outputs.
The same output engine can send a stream to another platform or write it to a file. Rules start those jobs automatically, a live buffer gives viewers instant rewind, and time-range parameters turn any available section into a clip.
Each input, output and processing step is its own process. They share memory and nothing else, so one of them falling over does not take the server with it.
MistServer is native code built on our own libraries rather than a deep dependency tree, and most of what it needs is linked in statically. Installing it is small, and it is as comfortable on an ARM board or an appliance as on a rack server.
Set up a family of streams once instead of adding them one at a time. Anything matching the pattern starts when it starts sending, and goes away when it stops.
SDI and NDI go in and out of MistServer directly. A feed can arrive on a cable and leave over IP, or the other way round, without a gateway box in between.
Combine and mix audio and video into a single output. Grids, picture in picture, alpha layers and fallbacks, all changed while the stream is running.
An encoder built on libav that runs inside MistServer and covers the modern formats, in hardware or in software. It handles raw pixels as well as encoded ones, which is what SDI work needs.
A player that passes codec data straight to the browser decoder. Much less to set up than WebRTC, latency in the same range, and it works over protocols WebRTC cannot use.
Everything MistServer does sits behind one JSON API, and the built-in web interface is a client of that same API. Anything you can click, you can script.
MistServer includes a balancer for a group of MistServer nodes. It points each viewer at a healthy node near them, and takes a failed one out of rotation without disconnecting anyone.
Hooks fire at ingest, playback, stream, recording and server lifecycle events. They can notify another system or block while your code approves, rejects, reroutes or rewrites what MistServer does next.
Inspect streams, tracks, viewers, inputs and outputs while they are live. Read recent statistics through the API, follow status over WebSocket, or scrape the built-in Prometheus endpoint for long-term monitoring.
Keep an MPEG transport stream byte-for-byte intact for a downstream system while an optional demuxer exposes its audio, video and metadata tracks to the rest of MistServer at the same time.
Serve on-demand media from local storage, HTTP or S3-compatible sources through the same delivery protocols as live streams. A folder can become a catalogue and a playlist can run as a linear channel.
Built and working their way to release. These reach the regular releases over the coming months, and we can hand you a build sooner if you need one.
A new management interface built around a canvas. Drop in a source, connect it to processing and destinations, and apply. MistServer writes the stream, process and push configuration for you, and every card shows whether it is live, idle or failing. You never have to open the rest of the interface.
Caching for ingest and for segmented output. On ingest it makes segmented inputs more reliable. On output it serves segments that have not changed rather than building them again for every connection.
Cameras and capture devices are found automatically and managed from MistServer, with control travelling alongside the video instead of through a separate system.
Models run against the stream on your own machine and return their results as metadata. The model pack covers detection, segmentation, pose, depth, faces, moderation, audio analysis, OCR and transcription.
Broader cue handling beyond today's splice-out insertion for push inputs and TS-based outputs. The next step is to carry and create SCTE-35 across more of the ingest and delivery path.
Send recordings and archives straight to object storage, with no folder-watching script sitting in between.
Multi-DRM for CMAF delivery, so protected content does not need a packaging chain of its own.
Work on the low-latency CMAF and DASH delivery paths.
MistServer requests and renews its own certificates, so there is no proxy in front of it purely for TLS.
Proper support for multi-program transport streams.
Cloud images and marketplace listings, so bringing up a node is a few clicks.
A better thumbnailer, and sprite sheets for scrub previews in your player.
Committed to, and further down the road.
Funded by an NLnet grant. Bringing on-demand up to the standard live already enjoys, including sharing assets between nodes so a VoD setup takes far less wiring.
Point MistServer at any provider or stack that speaks the OpenAI API, for teams who would rather not run models on their own hardware.
A shared protocol that gives a CDN enough understanding of a stream to relay it. Publish once, and relays fan it out.
Ready when you are
If something here lines up with what you are working on, we would like to hear about it.