Skip to content

Use cases

Five instances, and what each one needs.

The same software, sized and configured five different ways.

One person, one channel

An independent creator

You publish on a schedule and your back catalogue is the asset. What you do not want is for that catalogue to sit inside a recommendation system you cannot see, on terms that change without notice.

So you run one 4 vCPU box, put a CDN in front of it, and embed the player on your own pages. Your channel is addressable over ActivityPub, so people can follow it from wherever they already read.

What matters here

Embeds, oEmbed, RSS and a sitemap
Your video plays on other people's pages and your archive stays indexable.
The bespoke player
Keyboard shortcuts, picture-in-picture and theatre mode on a page you control.
ActivityPub and ATProto, per channel
Followers reach you over protocols rather than over an account you rent.
Resumable uploads
A 90-minute master over a domestic connection resumes instead of restarting.

SizingSmall, private: 4 vCPU, 8 GB, 160 GB — around $63 a month.Check the arithmetic →

Read the quickstart →

Many uploaders, one owner

A community or club

A few dozen members upload and a few hundred watch. The technical part is not the hard part — the hard part is that you are the person who answers for what appears on the instance.

Registration approval keeps signups behind a human. Per-user quotas stop one enthusiastic uploader from filling the disk, and ClamAV scanning is a flag rather than a project.

What matters here

Registration approval
New accounts wait for a person to let them in.
Reports and moderation
A reporting path for members and a queue for acting on it.
Per-user quotas
Storage limits per account, set before they matter rather than after.
Optional ClamAV scanning
Uploads scanned on the way in. Costs about 2 GB of RAM.

SizingSmall, private with headroom: 4 vCPU and 8 GB, plus 2 GB if you enable ClamAV.Check the arithmetic →

Moderation, from day two →

Several channels, live and archived

A newsroom or podcast network

You run several channels under one instance, and some of what you publish is live. A council meeting, a launch, a recorded interview embargoed until Thursday — each has a different audience and a different moment.

RTMP ingest takes the stream, privacy-gated HLS controls who can watch it while it runs, and the replay becomes a VOD without a second workflow. Embargoed material sits behind a scoped playback token rather than an unlisted URL.

What matters here

RTMP live with replay to VOD
The stream and the archive are one pipeline, not two.
Password-protected videos
An embargo that survives somebody pasting the link into a group chat.
Channel auto-sync mirroring
Keep a mirrored channel current without a manual re-upload.
Runtime-editable instance settings
Change instance behaviour during an event without a redeploy.

SizingPublic launch: 8 vCPU, 16 GB, 160 GB — around $168 a month.Check the arithmetic →

The pipeline, live to VOD →

Hundreds of talks, kept legible

A course or conference archive

You have four years of talks and every one of them has to stay findable. Nobody browses an archive; they search it, half-remember the title, and give up if the first result is wrong.

Hybrid full-text and trigram search with typo-tolerant autosuggest is the difference between finding a 2023 keynote and not. Whisper captions make the spoken content searchable as well as accessible.

What matters here

Whisper captions and WebVTT
Generated captions you can correct, on hundreds of hours of talks.
Chapters and storyboards
A long recording becomes navigable instead of a single scrubbing bar.
Hybrid search with autosuggest
Full-text and trigram together, so a wrong spelling still lands.
Accessibility gates in CI
Accessibility failures fail the frontend build — axe gates serious and critical findings — so the obligation is checkable.

SizingSmall, private to start, moving to public launch as the catalogue and concurrent viewers grow.Check the arithmetic →

How search works →

Many channels, a team behind them

An organisation outgrowing one box

Your catalogue and audience no longer fit the launch profile: encoding queues back up on release days, and one machine is a single point of failure that you answer for.

The image you installed on day one splits into roles — api replicas behind a load balancer, worker hosts for encoding — with object storage as the canonical store and a CDN with purge in front of public media. What is not built yet — multi-CDN steering, studio DRM, multi-region — is written down in a public roadmap with decision records, so you can judge the distance before you commit.

What matters here

Role-split deployment
One image, split into api and worker by a boot variable. No fork, and no separate enterprise build to buy.
Durable Postgres queues
Adding worker hosts adds encode throughput without double-processing — soak-tested with a deliberate counterfactual.
One-release rollback, enforced by CI
A schema-compat gate proves the previous release runs on the new schema before anything ships, and vidra update arms the rollback.
Zero-downtime storage migration
Local to S3 with hash verification at every step; rollback is one settings change.

SizingBeyond the published profiles. Validated at two api replicas and three workers on one PostgreSQL; a single-region deployment is the honest ceiling today, and live streaming stays on one host.Check the arithmetic →

The production deployment →

Start with one command.

One command. 4 vCPU. About $63 a month.

Install