Fully Managed
​Mirotalk SFU
as a Service

The constraint. Solicitor-client privilege is jurisdiction-specific. The Law Society of Ontario, the Bar Council of England and Wales, and the German BRAO all caution against client-confidential information transiting US-based conferencing vendors without an enforceable processor agreement and matching data-residency posture. The plain reading for a small firm: keep client video on infrastructure the firm controls.

What we set differently for law firms. The instance deploys to the region your bar association expects — Canada, UK, Germany, Australia, or any of the 21 locations. Room password and lobby mode are on by default. Host authentication routes through JWT bound to your case-management system so the room URL is one-time and tied to the file number. Recording is disabled at the platform level unless the retainer permits it, and when permitted, recordings stay on the backed-up volume in the same region as the call.

Workflow example. Client books a consultation via your booking platform; the booking webhook calls the MiroTalk SFU REST API; the API mints a JWT-signed room URL valid for that file number; the client receives the URL with their appointment confirmation; the room opens with the lobby armed. Around 6–10 client video consultations per day fit comfortably on a base €9 instance.

The constraint. Schools handling students under 16 fall under EU and UK data-protection rules requiring verifiable parental consent for any third-party processor that touches a student's video or voice, with equivalent rules under COPPA in the US and PIPEDA in Canada. Most commercial conferencing vendors satisfy the rule with a data-processing agreement, but the operational burden of vetting that agreement falls on the school. A self-hosted instance puts the school in the processor seat.

What we set differently for schools. Lobby mode is on by default so instructors admit students one by one — no anonymous joiners. The instance embeds via iframe into the LMS so the lesson page is the launcher and there's no separate account for students to register. Whiteboard and real-time polls are enabled. Recording policy is per-class and toggleable, so the school can record listening exercises and skip recording role-plays.

Workflow example. 1 instructor plus 6–12 students per class, with whiteboard used for grammar drills, YouTube embeds for listening exercises, and the integrated poll for comprehension checks. Average class duration 45–60 minutes. A school running 8 instructors and ~25 simultaneous classes at peak sits in the 50-participant range that a base €9 instance handles before needing a vCPU bump.

The constraint. Telehealth carries strict residency and processing rules — NHS Digital DSPT for UK GP practices, §75c SGB V for German doctors, PHIPA or PIPEDA for Canadian clinics, and equivalent US patient-privacy regulations for any practice with US patients. The common requirement: patient video does not transit a US commercial conferencing vendor without specific contractual protection, and the practice retains the right to audit the processor.

What we set differently for healthcare. The instance deploys into the region your regulator expects. Host authentication routes through OIDC bound to the practice's identity provider, so a consulting locum's access ends when the day ends. Per-room passwords are mandatory. Mediasoup worker count is tuned for one-to-one consultation latency rather than large-room fan-out — patient sees the doctor with under 100 ms one-way on a same-region instance. Recording, when enabled, stays on the regional storage volume and is included in the daily off-site backup of the same region.

Workflow example. 1 doctor running 6–10 patient consultations per day, each under 30 minutes, with the room URL sent via the practice's booking platform. A small practice of 3 doctors sharing one instance fits on a base €9 plan; a 10-doctor clinic typically needs a 2-vCPU upgrade once concurrent consultations cross 30.

The constraint. The numbers stop working once the agency, plus rotating freelancers, plus clients, cross 10–15 logins per month. Per-seat conferencing pricing scales with the team plus everyone the team talks to, and the brand experience of joining a client meeting on a third-party conferencing URL is not great. Agencies want their domain on the calendar invite.

What we set differently for agencies. The instance deploys to a custom subdomain on the agency's domain — meet.agency.com — with auto-renewed SSL. Branding overlay applies the agency's logo and palette and survives MiroTalk upstream updates. The integrated RTMP broadcaster wires to the agency's YouTube Live or Twitch destination, so a product launch streams from the same room the team uses for stand-ups, with no second tool. Persistent rooms get one URL per active client account.

Workflow example. Daily 15-minute internal stand-up; client-specific persistent rooms; ad-hoc room generator on the team intranet; one or two recorded creative reviews per week stored on the backed-up volume; occasional webinar to a public RTMP destination. A 5–10 person agency runs comfortably on a base €9 plan unless the webinars regularly cross 50 live attendees.

The constraint. Engineering teams need a conferencing tool that does not lock them out of pair-programming, screen review, and incident huddles when a vendor's free tier hits a time cap or the corporate SSO expires at the worst possible moment. Self-hosting is technically appealing; the operational cost of doing it well is what makes managed the better trade.

What we set differently for engineering teams. OIDC routes to the same identity provider that gates GitHub, the CI dashboard, and the on-call rotation — usually Okta, Authentik, Keycloak, or Microsoft Entra. The mediasoup port range is pinned so it does not collide with other services on the host. The REST API is wired into the team's incident-response runbook so a P1 alert auto-creates a bridge room and posts the URL to Slack.

Workflow example. Persistent rooms for daily stand-up, weekly architecture review, and the on-call bridge. Ad-hoc rooms minted from the runbook during incidents. Screen sharing for pair-programming and code review. A 10-engineer team usually fits on a base €9 plan; larger orgs with concurrent stand-ups and review sessions step up to 2 vCPU + 2 GB RAM.