Analytics cookies only (Google Analytics, PostHog). No ad networks, no cross-site tracking. See our privacy policy.
upstream: github.com/nextcloud/spreed (opens in new tab) · github.com/jitsi/jitsi-meet (opens in new tab)
Pick Nextcloud Talk
If you already run Nextcloud and want chat and calls tied to its users, with small calls and a TURN server you add yourself (~229 MB idle RAM).
Pick Jitsi Meet
If you want link-based meetings for outside guests with a bridge of its own, and can open ports and give it 2 GB or more (~336 MB idle RAM).
Nextcloud Talk and Jitsi Meet both give you self-hosted video calls in the browser, but they sit in different places. Talk is an app inside Nextcloud: chat, calls and screen sharing tied to your Nextcloud users and files. Jitsi Meet is a standalone conferencing server where guests join from a link without an account. The question that decides capacity is how calls are routed, which differs between them.
Talk's README calls it "A video & audio conferencing app for Nextcloud", with chat, private, group, public and password-protected calls, federated chats, screen sharing and integration with other Nextcloud apps. It installs from the Nextcloud app store, so it needs a Nextcloud server underneath.
Jitsi Meet's README describes "a set of Open Source projects" for deploying video conferencing, with HD audio and video, content sharing, chat, polls, virtual backgrounds, mobile apps and web and native SDKs. Our catalog notes the Docker setup runs the web frontend, Prosody, Jicofo and the Jitsi Videobridge as separate containers.
Talk's admin docs say it "works peer to peer": each participant sends a stream to
every other participant. Video is about 1 Mbit/s per stream, so bandwidth grows
with participants. The docs say a call with 20 people without video "could be doable"
on a fast network, and the TURN page says more than 5-10 users in the same call need
the Talk High Performance Back-end, a "partner
product" with a Selective Forwarding Unit that "typically scales to 30-50 or even
more active participants". The same page sends enterprise users to Nextcloud's sales team
for access. The signaling server is also published as AGPL-3.0 code with Docker images
(strukturag/nextcloud-spreed-signaling) and uses a Janus gateway. Either way it is
separate infrastructure to run or buy, not a setting in the app.
A TURN server is a separate matter. Talk's docs say "A TURN server running on port 443 (or 80) is required in almost all scenarios", and that it may be needed even with the High Performance Back-end. They also say HTTPS is required for WebRTC. The TURN page is softer for one case: "within one local network" it says nothing else should be required besides the app. Our Talk install has neither TURN nor HTTPS, and browsers block camera and microphone off localhost without HTTPS, so treat it as a test install until you add TLS.
Jitsi's Docker setup includes its own Videobridge to carry media. Our snippet sets the bridge's advertised
public IP and needs 10000/udp open for media. We did not test large meetings and
the upstream pages we read give no participant limit, so we state none.
Talk uses your Nextcloud accounts and can generate public links. Jitsi's
Docker docs describe optional authentication (ENABLE_AUTH, with internal, JWT or
LDAP modes) and a separate guest setting, so you can require logins to create
rooms while letting guests join. Talk is the better fit when your users and files
already live in Nextcloud; Jitsi is the better fit for ad-hoc meetings with people
outside your organisation.
Talk is AGPL-3.0 and Jitsi Meet is Apache-2.0. Both are open source. Talk's High Performance Back-end is a separate deployment: Nextcloud sells access, and the signaling server is published under AGPL-3.0. The core app is not gated.
Jitsi's handbook says RAM is "usually suggested to get 8 GB", that "For small meetings you can get away with 4 GB", and that for "very small meetings you can try to use 2 GB". Our 2 GB catalog figure is that last number. Talk's server requirements page gives no RAM figure, so our 1 GB is our estimate. In our install check (GCP e2-standard-2, Ubuntu 26.04), idle at first boot with the default install snippet and no calls, Talk used ~229 MB of RAM and ~3.5 GB of disk (November 2026), and Jitsi used ~336 MB of RAM and ~1.7 GB of disk (September 2026). The readings are from different months. Talk's disk figure covers the Nextcloud and PostgreSQL images too, and RAM at rest says nothing about RAM during a call.
Both are 3 out of 5. Talk is a Nextcloud plus PostgreSQL compose file followed
by an occ app:install spreed step. Jitsi needs a domain, ports 80 and 443 plus
10000/udp, and the official multi-container setup behind Caddy.
Choose Nextcloud Talk if you already run Nextcloud, want chat and calls tied to its users and files, and your calls are small or you will add a TURN server.
Choose Jitsi Meet if you want meetings that guests join from a link, a bridge that handles media in group calls, and do not need Nextcloud.
Both installs are verified on Ubuntu 26.04; the files are on the Nextcloud Talk and Jitsi Meet pages. Browsers block camera and microphone off localhost without HTTPS, so point a domain at the server and add TLS for either.
Nextcloud Talk lists a lower minimum: 1024 MB vs 2048 MB for Jitsi Meet.
No. Nextcloud Talk is AGPL-3.0, Jitsi Meet is Apache-2.0.
Nextcloud Talk if you already run Nextcloud and want chat and calls tied to its users (~229 MB idle RAM in our first-boot check, ~3.5 GB disk with Nextcloud). Jitsi Meet for link-based meetings with guests and its own videobridge (~336 MB idle, ~1.7 GB disk).
Its docs say a TURN server on port 443 or 80 is required in almost all scenarios, and may be needed even with the High Performance Back-end. Its TURN page says that within one local network nothing else should be required besides the app, but HTTPS is still needed for WebRTC. Our install has neither TURN nor HTTPS.
Talk works peer to peer, so bandwidth grows with each participant. The docs say a call with 20 people without video could be doable on a fast network, that more than 5-10 users in one call need the High Performance Back-end, and that it typically scales to 30-50 or more active participants. Nextcloud sells access as a partner product; the signaling server is also published under AGPL-3.0.
Yes in the default setup. Jitsi's Docker docs describe optional authentication (ENABLE_AUTH) with a separate guest setting, so you can require logins to create rooms and still let guests join.
Paid link — we earn a commission if you shop through it.
A whiteboard-and-docs workspace on a plain Compose file vs. native apps on a commercial server.
A full identity platform vs. a passkey-only sign-in box.
Two modern IdPs, built for different jobs.