Live streaming and restream

Self-host RTMP and SRT ingest, multi-platform restream, and live-to-VOD capture on Quake AI Compute. You run NGINX-RTMP, SRS, or similar ingest software on VMs you own; Quake AI provides the compute, a stable floating IP for encoder pushes, networking, and object storage for captured video.

What this is for#

Streamers, broadcasters, and platforms that need control over ingest run their own RTMP/SRT endpoint instead of a managed live-video service. Quake AI provides a stable floating IP for encoder pushes, CPU compute for the ingest and restream process, and object storage for live-to-VOD capture; you operate the ingest software and restream configuration. The outcome is a self-hosted ingest and fan-out path with flat egress, deployable from a validated OpenTofu template and its companion tutorial.

Reference architecture#

Download diagram: SVG, PNG, and PDF.

The base is the live ingest and restream template: a floating IP, an ingest VM, and encoder-scoped security groups. The dashed boundary is the archive variation, which adds an Object Storage bucket when you want live-to-VOD capture alongside the live fan-out.

1. Encoders. OBS, hardware encoders, or an upstream source publish a single RTMP or SRT stream to a stable public address.

2. Floating IP. A floating IP gives encoders a stable ingest address on port 1935/tcp (RTMP) and the SRT UDP port if enabled. Security groups scope the ingest ports to known encoder CIDRs.

3. Ingest VM. An ingest process (NGINX-RTMP, SRS, or similar) runs on a Compute instance, receives the stream, and fans it out to downstream destinations. The live ingest and restream template provisions the instance, floating IP, security groups, and restream configuration.

4. Restream targets. The ingest VM pushes copies of the stream to external platforms (YouTube, Twitch, custom RTMP origins) listed in the template's restream_targets.

5. Live-to-VOD capture (archive variation). The archive variation adds an S3-compatible Object Storage bucket, and the ingest process records the live stream to it for later on-demand playback or archiving. Leave it off for live-only fan-out.

Services involved#

Get started#

• Live RTMP/SRT ingest and restream template: OpenTofu pattern with a public ingest floating IP, encoder-scoped security groups, and configurable restream targets.
• Deploy the live ingest and restream template: end-to-end OpenTofu walkthrough from apply through OBS publish and teardown.
• Simple VM template: single-instance layout when you install and configure ingest software yourself.
• Deploy the Simple VM template: step-by-step walkthrough for a single Compute instance with a floating IP.
• Floating IPs: stable public address for encoder pushes to RTMP or SRT.
• OpenTofu template library: browse validated IaC starting points for the layouts above.

Estimate the cost#

Migrating an existing live streaming / restream workload?#

Move a live ingest and multi-platform restream stack that already runs on AWS EC2 and VPC, Azure VMs, Google Compute Engine, DigitalOcean Droplets, or another provider's VM fleet. The outcome is the same encoder workflow and downstream restream destinations on Quake AI compute and networking, with ingest URLs and platform outputs cut over in a controlled order.

Follow this cutover path. Each step links an existing migration page; this section composes those pages into a workload-shaped sequence rather than duplicating their steps.

1. Map your source provider. Start with the concept-translation page for your current cloud: Coming from AWS, Coming from Azure, Coming from GCP, Coming from DigitalOcean, Coming from Hetzner, Coming from Linode, or Coming from Vultr.

2. Stand up the target shape on Quake AI. Pick the template that matches how you run ingest after migration: Live RTMP/SRT ingest and restream for dedicated ingest and fan-out, Simple VM for a single ingest host, or edge reverse proxy when HLS or HTTP playback needs a dedicated HTTPS origin.

3. Move compute workloads. Rebuild or migrate ingest and restream workers with Migrate from EC2 (or the matching compute migration page for your source provider).

4. Recreate network topology and security rules. Translate VPCs, subnets, and firewall rules that scope encoder and admin CIDRs with Migrate from AWS VPC (or the matching network migration page for your source provider).

5. Validate on the target stack. Apply the template or migrated instance, set encoder_cidr and restream_targets, and run a test publish from a staging encoder before you touch production encoders.

Workload-specific cutover callouts#

• Ingest endpoint cutover. Production encoders (OBS, hardware encoders, or upstream CDNs) must use the Quake AI ingest URL, typically the rtmp_publish_url output or the new floating IP on port 1935/tcp (and SRT UDP if enabled). Update encoder profiles during an off-air window so viewers never see a half-switched path.
• Restream target reconfig. Downstream RTMP URLs in restream_targets must match what each platform (YouTube, Twitch, custom origins) expects after migration. Re-enter stream keys and verify each destination accepts a test push before you go live on the new host.
• Low-downtime switch during off-air window. Schedule the production encoder switch when you are not live. Keep the source ingest host running until all restream destinations show healthy bitrates on Quake AI; dual-publish only if your encoder supports it and your platforms allow duplicate stream keys.
• Rollback. Preserve the source ingest instance and its security group rules until error rates and restream health stabilize. Roll back by switching encoders back to the old ingest URL and restoring previous restream_targets if probes or platform dashboards fail.

Considerations and limits#

• No managed live video. Quake AI does not provide managed ingest, transcoding, or packaging. You install and operate the ingest software (NGINX-RTMP, SRS, or similar) on Compute under the shared responsibility model.
• CPU-only transcode. Compute is AMD EPYC with no GPU option (compute FAQ). Heavy real-time transcode at scale is CPU-bound; size the ingest VM for passthrough fan-out plus any software transcode you run.
• Flat egress. Quake AI applies a no-egress-fee policy for outbound transfer, which suits restream fan-out to multiple downstream platforms.
• Three US regions. All current regions are in the United States. Geographically distributed ingest needs additional points of presence you operate.
• Compliance posture. Quake AI holds SOC 2 Type I and Type II attestations and SOC 3. See Compliance and certifications for the platform scope.