The first version took three days. Then came roughly two weeks of making it fast. In an account of Stream’s switch from Python to Go, Thierry Schellenbach described a ranking feature whose tidy development schedule concealed a much less tidy optimization schedule. The later Go implementation took about four days, without that extra tuning. He acknowledged the comparison’s limitations: the problem was better understood the second time. Still, the calendar had delivered a useful correction. Shipping code and owning it were different jobs.
- Stream sells chat, feeds, video, and moderation infrastructure to app builders.
- The customer controls the experience; Stream runs much of the machinery underneath.
- The buying decision turns on maintenance, usage costs, and how much control a team needs.
That distinction explains Stream better than a catalogue of features. A message composer occupies a few centimeters of a phone screen. Behind it are connections, delivery, permissions, storage, notifications, and the awkward business of reconnecting when someone walks into an elevator. Every app team can begin this project. The question is how much of its future it wants to spend finishing it.
The fashion network behind the infrastructure
Before Stream, Schellenbach founded Fashiolista, a fashion social network. Tommaso Barbugli was its lead engineer. Together they worked on the open-source Stream Framework, a library for activity feeds. These were people who had encountered the problem from inside a social product: somebody posts something, somebody else follows them, and the right update must appear in the right place.
They founded Stream in 2015, joined Techstars NYC, and subsequently established its headquarters in Boulder. Amsterdam remained part of the company’s geography. The change in business was consequential: instead of asking people to join another social destination, they sold other builders the infrastructure for their own destinations. The customer’s brand could occupy the screen. Stream could occupy the plumbing.
Its expertise became the repeated, difficult work underneath familiar interactions: distributing updates, keeping conversations synchronized, and supporting applications across devices. That is an attractive position in software. A buyer can care intensely about a feature without wanting the organization required to operate it.
The customer keeps the reason to exist
Consider Healthline. Its condition-focused community apps needed feeds, but its distinctive job was serving people with particular health concerns. In Stream’s published customer account, Healthline considered building the social infrastructure internally. Using Stream let its product team put more attention into each community’s needs. The feed mattered because it carried the experience; it was not the whole experience.
The same division of labor appears across very different businesses. Stream names Strava, Patreon, Nextdoor, IBM, and Adobe among its customers. It reports reaching more than a billion end users through customer applications. That figure describes reach through other products, rather than a billion people opening Stream accounts.
“The API economy creates a golden age for engineers and product teams.”Thierry Schellenbach · 2021 financing announcement
The toolkit now covers several parts of that experience. Chat supplies messaging and interface components. Activity Feeds handles timelines and social updates. Video adds calls, audio rooms, and livestreams. Moderation adds content screening and review workflows. Developers can use prepared interface components or work closer to the underlying APIs, depending on how much they need to customize.
Sendbird and CometChat compete for messaging work; Agora, Twilio, and Vonage overlap in communications. Another alternative is an internal engineering project. Stream’s argument combines backend infrastructure, client SDKs, and customizable interfaces across several related products. Whether that combination wins depends on the buyer’s platforms, traffic, requirements, and willingness to depend on a vendor.
Count the people. Then count the minutes.
The bargain has a visible price. Stream’s published Chat Start plan lists $399 a month when billed annually, or $499 on monthly billing, including 10,000 monthly active users and 500 concurrent connections. Those are different quantities. A popular app can have many people visit during a month while relatively few are connected simultaneously.
Video uses another clock. Four people on a thirty-minute call consume 120 participant minutes. Received video resolution also affects the bill, while recording and transcription can add charges. A product manager comparing vendors needs a model of actual behavior: how many people arrive, how long they stay, and what they receive. A headline subscription price is only the beginning of that calculation.
Participant minutes, before resolution and optional services.
Free developer tiers lower the cost of trying the idea. A qualifying Maker program supports small teams under specified revenue and funding thresholds. But integration still takes work. Stream’s migration guidance describes exporting data and replacing application components. Buying infrastructure creates a dependency that a team should understand before embedding it deeply.
For a team with unusual deployment requirements, an existing communications platform, or economics that favor operating its own stack, the calculation can point elsewhere. The sensible comparison includes integration and exit costs alongside development time. A feature that looks ordinary to a user may be central to the business’s technical advantage. That team has a stronger reason to retain direct control than a publisher simply trying to add a conversation beside its content.
Reliability is also something buyers purchase in degrees. Stream offers an enterprise package with a 99.999 percent uptime SLA and round-the-clock support. An SLA is a contractual commitment with remedies, rather than proof that an application will never go down. Customers still need to examine their own implementation. A dependable service cannot correct every permission mistake or decide which messages belong in a community.
There was capital behind the expansion. A $15 million Series A in August 2020 preceded a $38 million Series B, led by Felicis Ventures, in March 2021. The latter announcement reported 517 percent growth in chat API revenue during 2020. That is a dated product-growth figure, not a disclosure of today’s company revenue.
A conversation with something that can see
Stream’s newer work extends the conversation to machines. Vision Agents is an open-source Python framework for live voice and video AI. Its 2026 releases added HTTP deployment support and more integrations. April’s v0.5 release also addressed resource leaks in long-running deployments. Even an AI demonstration eventually meets the ordinary demands of running software.
The company’s public ten-week Go onboarding curriculum offers another clue to its habits: databases, performance, testing, and review get explicit attention. For readers building products, the useful lesson is to budget for the full life of a feature. Rent the common machinery when that frees the team to do something customers will notice. Keep the responsibilities the API cannot take away: deciding who may speak, what the community needs, and whether the conversation deserves to happen at all.