Jitsi Meet vs Nextcloud Talk

Jitsi Meet

The videobridge relays rather than mixes, and the engineering numbers stand behind scale — 1056 streams at roughly 20% CPU on a single rented server, with Last-N trimming for large rooms. Control is the weak flank: on the public service all users are moderators and the moderation controls are documented as "soft, and for everyone", and I found no public information on waiting rooms, breakout rooms or per-meeting participant capacity.

Nextcloud Talk

Breakout rooms and webinars-with-lobby exist, and published scaling is candid — 6-10 participants for a private setup, thousands only at enterprise — but the evidence is silent on screen sharing, mute-all or any host moderation, and says nothing about behaviour on a constrained connection beyond admitting video caps you at 20. Between the 3 and 5 anchors: breakout rooms without the moderation and screen-sharing basics that rubric level 5 demands.

Jitsi Meet

The mechanism is documented honestly rather than waved at: hop-by-hop DTLS-SRTP whose outer layer the bridge strips in multiparty meetings, and optional end-to-end encryption whose second layer is "never removed" at the server, with the trade-offs named since chat and polls sit outside it. I hold it below the top band because the key is a shared passphrase every participant types rather than per-device keys or identity verification, and getting in is a link plus an optional password; I found no public information on a lobby or SSO-gated joining.

Nextcloud Talk

"Multi-layered encryption" is a marketing phrase, not a mechanism: no end-to-end statement for meetings, no key handling, and signaling can run through a High Performance Backend hosted by unnamed partners. The only concrete cryptography I can verify is push notifications encrypted and re-signed by the vendor's proxy, with a lobby available only for webinars that buy the HPB. Nothing here tells me what the server operator or the hosted HPB can see of the media.

Jitsi Meet

Browser joining without an install and guests without accounts are the design centre, with mobile apps on the App Store, Google Play and F-Droid, plus a dial-in information endpoint and speech-to-text transcription. The pages stop short of reach as a promise, though: we found no public information on dial-in numbers, accessibility conformance, keyboard and screen-reader testing, or low-bandwidth behaviour beyond Last-N stream trimming.

Nextcloud Talk

Web, desktop and mobile are covered, live transcription with translation exists, and there is a real telephone path — but dial-in requires the separately deployed HPB and no numbers or rates are listed. There is no accessibility statement, no keyboard or screen-reader evidence, and no documented low-bandwidth behaviour anywhere in the evidence; silence here is information.

Jitsi Meet

The default is discard: rooms are ephemeral, chat and speaker statistics are destroyed when the meeting ends, and a cloud recording sits on the servers at most 24 hours before upload to the destination the organiser chose. Governance is where the pages go quiet — we found no public information on who may start a recording, participant notification or consent, organisation-wide recording policy, storage region, or an access log for stored recordings.

Nextcloud Talk

There is a recording-consent prompt in meetings and self-hosting puts the recording on infrastructure the customer controls, which is a decent starting point. But the evidence evidences no admin policy over who may record, no retention or deletion rule for recordings or the AI transcripts, and no access logging — message expiration is a chat feature, not recording governance, and nothing here is works-council switchable.

Jitsi Meet

Deployment freedom is the strong card: a stack that is 100% open source under Apache 2.0, self-hostable in about fifteen minutes via Docker, controllable over XMPP and a REST interface, embeddable through the external API, with tenant-named meeting URLs. The conventional stack side is thinner — we found no public information on calendar integration with the major suites, SAML or OIDC single sign-on, SCIM provisioning or webhooks.

Nextcloud Talk

Deployment is genuinely the customer's: fully on-premises, airgap-capable, open source, with webhook bots, a SIP bridge and a Pexip gateway for room systems and an Outlook add-in. But the evidence shows no SAML/OIDC SSO, no SCIM provisioning and no SDK — for a self-hosted stack, the absence of documented identity federation is a real gap.

Jitsi Meet

The independent sovereignty assessment sits at the bottom of the scale and I weigh that heavily here: the contracting party is 8x8, Inc. of Campbell, California, disputes are litigated in New York courts, and analytics on the public service run to Amplitude, Datadog and Crashlytics. We found no public information on where media for the vendor-run service is processed, on any EU contracting entity, or on a complete subprocessor list; self-hosting would move the question onto the customer's own infrastructure, but that is a deployment the buyer builds, not a guarantee these pages evidence.

Nextcloud Talk

The contracting entity is firmly German — Nextcloud GmbH, register HRB 227086, Stuttgart, German VAT ID — and the self-hosted/airgap model means media locality is largely the customer's own decision, which is why I score above the 5 anchor. But nothing evidences where vendor-operated pieces land: the STUN server Nextcloud itself hosts, partner-hosted HPBs with unnamed partners, and Matomo cloud for analytics; there is no product subprocessor list, so the chain is partly a black box.

Jitsi Meet

The headline is public and simple — the demo service is free, "all day, every day", with no account — but the same pages say it is not meant for use in production applications, and we found no public pricing for the managed Jitsi-as-a-Service offering a production buyer would actually run on. The terms also reserve 8x8's right to require a transition to a paid service at its sole discretion above a usage limit we found no public figure for, so no annual invoice is computable from public pages.

Nextcloud Talk

Per-user prices are public for three tiers with volume breaks, in EUR per year net of VAT, from 100 users — better than a headline, but capacity limits per plan are absent, minimum term is unstated, and dial-in is entirely unpriced because it depends on an HPB whose hosted cost never appears. Recording storage is likewise unmentioned; the annual invoice for what a buyer actually needs is not computable from these pages.