This page is alive. The uptime ledger ticks per day. The postmortems are linked, not summarised. The audits have dates and auditor names. If we ever stop publishing them on time, you get to hold us to it — that's what makes the claim mean anything.
The first vendor we evaluated had a SOC 2 logo and a marketing page about reliability. MPEGFLOW sent us the link to their last three postmortemsin the first reply. That's how we knew which one was actually built for us.
— HEAD OF PROCUREMENT · TIER-1 BROADCASTERSeven renditions, two DRM schemes, four caption tracks, twelve audio variants — packaged as an IETF-spec HLS master playlist in 1 minute 12 seconds.
Your workflow declares a transcode preset. A worker runs that preset against your source — one FFmpeg invocation, one job, one set of outputs. Below: a 5-rung HLS ladder, the actual shape of those outputs over a 9-second window of one real source.
One coordinator, 96 encoders, 14 regions. When a region goes dark, your viewers don't notice — the segments just keep arriving. And throughput stays linear: every encoder you add is another 14 seg/s, exactly.
That counter on the previous chapter doesn't go up because we batch jobs into shared processes and pray. It goes up because every job runs in its own worker — one pod, one process, no shared state. When a 4K render crashes, exactly one job is affected. The other eleven keep shipping.
The day a worker OOM'd halfway through a Champions League render, our previous platform took the whole pool with it. Here, the other 23 jobs didn't even notice.Crash isolation isn't an architecture diagram for us — it's a Friday-night rollback we never had to do.
— HEAD OF BROADCAST OPS · HORIZON MEDIAEvery job ships its FFmpeg progress and stderr to the API in real time. Every state change emits a domain event. Three subscribers fan out automatically: a row in the audit log, a WebSocket message to your dashboard, and an HMAC-signed webhook to whatever you're paging.
A node-graph editor for video pipelines — every pipeline is a directed acyclic graph (DAG) of callable operations. 48+ native nodes — sources, probes, transforms, packagers, AI, quality gates, delivery, control flow, notifications — plus every FFmpeg CLI option exposed as a callable node. Connect them with typed edges. Submit jobs against the graph. The graph serializes to YAML or JSON, versions in git, and IS the contract.
The workflow you saw in the last chapter ships as a single declarative file. Here is every surface that file ever has to touch — the SDKs that send it, the events that stream back, and the timer that says how fast a developer who's never seen us before gets a real frame on screen.
$ curl -X POST https://api.mpegflow.com/v1/jobs \
-H "Authorization: Bearer $MPEGFLOW_KEY" \
-H "Content-Type: application/yaml" \
--data-binary @workflow.yaml
{
"event": "job.queued",
"id": "job_8c4e9a7e2af",
"ts": "2025-04-26T14:32:08.412Z",
"queue_position": 7,
"queue_depth": 184
}We integrated MPEGFLOW the same morning the contract was signed. Read the docs, wrote the YAML, watched the first frame land at 3 minutes 51 seconds. No sales call between the curl and the manifest.
— PRINCIPAL ENGINEER · OTT PLATFORMThe same control plane runs on shared SaaS, your dedicated cluster, your VPC, your on-prem k8s, and an air-gapped vault. Pick the residency that survives your procurement review — the workflow you saw in chapter 06 doesn't change.
Our security review said no SaaS, ever. We deployed MPEGFLOW into our own VPC in the morning, ran the same workflow YAML as everyone else by lunchtime, and shipped to air the next week. The only thing the procurement team had to write was the cheque.
— CTO · NATIONAL BROADCASTEREvery artefact your security, legal, and procurement teams need to sign — already current, already shared. SOC 2 Type II, ISO 27001, HIPAA technical safeguards, DPA, BAA, BYOK addendum, named TAM, 24/7 escalation. Not a "request for assets" email chain. A binder.
We've onboarded fifteen vendors in the last year. MPEGFLOW is the only one whose security team handed us a completed binder before we asked the second question. Procurement closed the file in five business days.
— HEAD OF INFOSEC · TIER-1 BROADCASTERFive ways to buy it. Same control plane underneath. No mystery meat — every rate is on the rate card, every tier is the same product.
MpegFlow is a video pipeline engine that models transcoding and streaming workflows as a directed acyclic graph (DAG). Each stage — ingest, encode, QC, package, deliver — runs on a fleet of FFmpeg workers with retries, webhooks, and a per-job audit trail. The same binary runs as managed SaaS or self-hosted in your own datacenter.
FFmpeg is the encoding workhorse — MpegFlow is the operational layer around it. We provide the queue, retry semantics by failure class, partial-success handling on ABR ladders, encoder-version pinning, multi-tenant security, and per-job audit trail. You bring the FFmpeg knowledge (codec choices, presets, QC rules); MpegFlow handles the orchestration that every video team otherwise rebuilds from scratch.
Engineering teams running video infrastructure at broadcast, OTT, archive, or premium VOD scale. Specifically: teams transcoding 10K+ minutes per month, dealing with ABR ladder generation, multi-rendition outputs, audit and compliance requirements, or self-hosted deployment needs. If your video volume is under 1K minutes/month or your workload is consumer UGC, managed services like Mux or Cloudflare Stream are usually a better fit.
Your data stays in your storage. MpegFlow workers receive coordinator-issued presigned URLs to read mezzanine assets and write outputs directly to your S3-compatible bucket (AWS S3, Cloudflare R2, GCS via interop, MinIO, on-prem object storage). Our control plane never relays your bytes; it stores only job metadata, audit logs, and pipeline definitions. EU data residency is supported via region-pinned coordinator deployment; self-hosted runs entirely in your own datacenter or VPC, with zero data leaving your perimeter.