What "open source" means in this category, precisely
Team chat's open source field is real, mature, and badly labeled. Vendors say "open source" for at least three structures that behave differently in production and in law, and the difference decides what you own. Structure one: the whole product is open source under an OSI license (Zulip under Apache-2.0; you can run every feature the project ships). Structure two: an open protocol with open implementations (Matrix, with the Element client under Apache-2.0; the openness is architectural, and federation is the point). Structure three: open core, where a community edition under an open license (Rocket.Chat's MIT core; Mattermost's open core model) coexists with paid editions holding enterprise features like advanced identity, compliance, and scale tooling. None of these is a scam; all three are misdescribed by roundups that print the words "open source" and a star rating. This audit prints licenses and boundaries instead. The sovereignty reading of much of this same bench is in the Europe guide, and the budget reading is in the free cap audit; this page is the license reading.
| Tool | License | Structure | Where the free edition ends |
|---|---|---|---|
| Zulip | Apache-2.0 | Fully open product | It does not; cloud is the paid convenience |
| Element / Matrix | Apache-2.0 | Open protocol and clients | Hosting and support are the paid layer |
| Rocket.Chat | MIT (community core) | Open core | Enterprise features sit in paid editions |
| Mattermost | Open core (AGPL and commercial mix) | Open core | Paid editions hold enterprise features; cloud prices null here |
| Nextcloud Talk | AGPL-3.0 (Nextcloud) | App inside open platform | The platform is open; your operations are the cost |
| Jitsi Meet | Apache-2.0 | Open meetings | Self-host free; hosted convenience is paid |
The fully open product: Zulip
Zulip is the cleanest license story in team chat: an Apache-2.0 project whose self-hosted edition is the product, with a cloud offering (Standard at $6.67 per user per month billed annually, $8 monthly) that sells hosting and convenience rather than unlocking features. Its topic-threaded model is the structural fix for chat's knowledge problem, argued fully in the Linux guide where its desktop client also stars. For an open source buyer, Zulip's grading is simple: the license permits use, modification, and redistribution including commercially, there is no edition cliff to audit, and the project's governance is a community-and-company hybrid with a public track record. If your requirement is "open source, full stop," this is the bench's reference point.
The open protocol: Element and Matrix
Matrix moves the openness down a layer: the protocol is an open standard, servers and clients are open source (Element under Apache-2.0), and organizations can run their own server, federate with others, and encrypt rooms end to end. The practical consequence is exit-proofing: no vendor can hold your network hostage because the network is not a vendor's. The costs are the ones federated systems always charge: someone operates a server, client polish varies, and features arrive at protocol speed rather than product speed. Element the company sells hosting and support, which is the honest paid layer: convenience, not capability. Public bodies across Europe run Matrix for exactly these properties; the jurisdiction grading is in the Europe guide.
The open core pair: Mattermost and Rocket.Chat
Mattermost and Rocket.Chat are excellent products with a boundary to audit. Both offer self-hostable community editions under open terms (Rocket.Chat's community core is MIT; Mattermost combines open licensing with commercial editions), and both reserve enterprise territory (advanced identity and access, compliance tooling, high-scale deployment features) for paid editions. This dataset carries Mattermost's paid cloud prices as honest nulls and Rocket.Chat's cloud prices likewise, because neither could be verified cleanly in this pass; the self-hosted community editions are the verifiable $0. The audit question for open core is specific: list the enterprise features your organization actually needs (SSO depth, retention controls, audit logs, clustering), and check each against the edition boundary in the vendor's own documentation before adopting. Teams that skip this step discover the boundary in production, where it is priced accordingly.
The platform plays: Nextcloud Talk and Jitsi Meet
Nextcloud Talk is chat and calls as an app inside Nextcloud (AGPL-3.0), which makes its open source question inseparable from its sovereignty question: run the platform and the data path is entirely yours. Jitsi Meet (Apache-2.0) covers the meetings slice: self-hosted video that pairs with any chat above, at the cost of operating the most failure-visible component yourself. Both belong in this audit because many "Teams alternative" requirements are really bundle requirements, and the open source answer to a bundle is a composition: Zulip or Mattermost for chat, Jitsi for meetings, Nextcloud for files. Compositions are more work and more yours; that trade is the entire philosophy of this bench, stated once and applicable everywhere.
