Mattermost

Here is a fact that should be more surprising than it is: an awful lot of the most sensitive conversations in the world - troop logistics, incident response at a bank, a defense contractor coordinating a launch - happen in a group chat. And group chat, as a category, is dominated by two products, Slack and Microsoft Teams, whose core promise is that you don't have to think about where any of it lives. Someone else runs the servers. Someone else holds the data. That's the entire pitch, and for most companies it's a good one.

Mattermost is a bet that for a meaningful minority of organizations, "someone else holds the data" is not a feature but a disqualifier. If you are an air force, or a central bank, or a company operating on an air-gapped network with no internet connection at all, the sentence "your messages are safely in our cloud" is closer to a threat than a comfort. Mattermost's answer is deceptively simple: here is the whole platform, it's open source, run it yourself, on your own hardware, inside your own walls, and never send us a single message.

The mechanics of the product look familiar. There are channels and direct messages and threads and file sharing and search - the grammar of modern team chat, which Slack largely invented and everyone else now speaks. There are voice calls and screen sharing. There is a mobile app. If you squint, it is a Slack that you host. But the interesting part is not what it copied; it's the constraint it accepted. By insisting the software must run anywhere - including on a laptop in a facility with no network, no telemetry, no phone-home - Mattermost gave up a lot of the conveniences that make cloud software easy to build and easy to sell. In exchange it got a customer base that genuinely cannot go anywhere else.

That customer base is unusual. Public references include the U.S. Air Force, NASA, Nasdaq, the European Parliament, SAP, and Samsung. Uber, in the way large engineering organizations sometimes do, didn't just deploy Mattermost - it built its own internal chat product, uChat, on top of it, migrated roughly 20,000 rooms onto it, and pushed more than a million messages a day through it. When your product is open source, your biggest customers stop being customers and start being co-builders, which is a very different relationship than the one Slack has with its accounts.

In 2011, Ian Tien and Corey Hulen started a studio called SpinPunch to build an HTML5 game engine. Like most studios building game engines in 2011, it did not become a wildly successful game company. But along the way the team built an internal messaging tool to run itself, got frustrated with the commercial chat product they'd been using, tried to switch, and discovered they were locked into a vendor's ecosystem in a way they didn't like.

So they repurposed their own in-game messaging code into a standalone chat app, and in 2015 they did something a little unusual for a company with no open-source track record: they gave it away. Tien has described open-sourcing Mattermost as basically an experiment - a test to see whether anyone cared. The response was a large, unsolicited pull request from a stranger and a wave of community traction that exceeded anything the founders expected. That reaction, more than any business plan, is what told them there was a company here.

It's a useful reminder that you often don't know what you've built until you let other people touch it. The game was the plan; the chat tool was the byproduct; the byproduct is now the thing that governments run on. Persistence beat brilliance, which is how these stories usually go when you look closely.

The economics are the classic open-core trade. Give away the engine, charge for the things a serious institution can't operate without: audit trails, compliance, access controls, support contracts, and now AI that runs behind the firewall. Third-party sources put revenue around $33 million in 2024, up from roughly $20 million the year before, on a team of about 160 people. That is not hypergrowth in the SaaS-headline sense, but it's the shape you'd expect from a company whose customers sign long, deliberate contracts and rarely churn because switching would mean re-architecting how a government talks to itself.

The funding history is similarly measured. Mattermost raised early rounds from Redpoint and S28 Capital, and in June 2019 Y Combinator Continuity led a $50 million Series B, joined by Battery Ventures and the existing investors - bringing total funding past $70 million. The valuation has never been made public, which is itself on-brand for a company whose entire pitch is that some information should stay inside the building.

The current wave of the company's story is about AI, but with a twist that follows directly from the founding constraint. Plenty of software now offers an AI assistant; the assistant lives in a vendor's cloud and talks to a model in another vendor's cloud, which is fine until your data is classified. Mattermost's pitch is agentic AI that runs behind the same firewall as everything else - able to plug into a sovereign model or a third-party LLM, act inside the channel where the work is happening, and pull grounded answers back out of the team's own archive.

In June 2025 the company packaged this as the Intelligent Mission Environment, aimed squarely at defense, government and critical-infrastructure teams, and joined the inaugural Oracle Defense Ecosystem to deliver it on sovereign cloud infrastructure. Earlier that year it announced intent to bring AI-powered collaboration to Microsoft Azure's Secret and Top Secret environments. By 2026, Mattermost Agents V2 turned agents into something a team can own and operate rather than a novelty in a sidebar.

It is worth noticing the timing. Data sovereignty spent years as a compliance checkbox and then, quite suddenly, became a headline. Mattermost had already spent a decade building for exactly that world - air-gapped, self-hosted, auditable. Sometimes the strategy is just patience, and being ready when the rest of the market decides your niche was the important one all along.