A Startup Built on a Chore Nobody Wanted
Here is a fact about software that is both extremely boring and extremely load-bearing: if you run an API, you have to give people SDKs. An SDK is the little library a developer installs so they can call your API in Python or TypeScript or Go without hand-rolling HTTP requests and squinting at your docs at midnight. Every API-first company needs them. And every API-first company quietly resents them, because SDKs are software that has to be written, tested, versioned, published, and then - this is the cruel part - rewritten every single time the API changes. Which is constantly.
Speakeasy is a San Francisco company whose entire premise is that this chore should not be a chore. Founded in 2022 by Sagar Batchu and Simon Yu, it makes a fairly bold claim about a fairly narrow problem: give us your OpenAPI specification - the machine-readable contract that describes what your API does - and we will generate production-grade, type-safe SDKs across ten-plus languages, publish them to the package managers, and regenerate them automatically every time your spec changes. You do the API. The SDKs become a byproduct.
This is the kind of idea that sounds obvious right up until you try to build it, at which point it becomes a nightmare of language idioms, edge cases, and the fact that TypeScript developers and Go developers want fundamentally different things and will both be annoyed. The bet Speakeasy made is that the annoyance is worth centralizing. If one company obsesses over generating genuinely idiomatic SDKs - not the stilted, robotic output that open-source code generators are famous for - then thousands of API teams never have to.
Two Founders and a Receipt for Their Frustration
The origin story is the good kind, which is to say it is a complaint. Batchu and Yu had lived the API developer-experience problem at both startups and large enterprises. They had felt, repeatedly, the specific pain of maintaining SDKs by hand - the pull requests, the drift between spec and library, the language they didn't have an expert for. When a chore annoys you at several jobs in a row, it stops being a chore and starts looking like a company. So they built the tool they wished had existed, and named it Speakeasy.
Investors, it turned out, recognized the pain too. In mid-2023 the company emerged having raised roughly $11 million in pre-seed and seed funding, including a $7.6 million round led by GV, Google's venture arm. The angel list read like a Rolodex of people who have personally suffered from bad developer tooling: the president of Datadog, the CEO of Cribl, the CTO of Attentive. Then, in October 2024, Speakeasy raised a $15 million Series A led by FPV Ventures, with GV and Quiet Capital following on. Total raised: around $26 million for a company selling what is, on paper, middleware.
What You Can Actually Do With It
The mechanics are refreshingly literal. You point Speakeasy at an OpenAPI spec. Its tooling validates and enhances that spec - because most specs, left to their own devices, are a little bit wrong - and then it emits SDKs in TypeScript, Python, Go, Java, C#, PHP, Ruby, and more, along with Terraform providers for the infrastructure crowd. The TypeScript SDKs ship with runtime type safety powered by Zod and a single runtime dependency, which is the sort of detail that makes engineers quietly nod. The whole thing runs as a standalone CLI binary, which means it slots into CI pipelines and works even in air-gapped enterprise environments where nothing is allowed to phone home.
The payoff is that the SDK stops being a project and becomes a build artifact. When your spec changes, the SDKs regenerate. When you cut a release, they publish. The customer roster suggests this actually works at scale: Vercel, Mistral AI, Clerk, Kong, Fivetran, Airbyte, and SolarWinds all lean on it. Mistral AI, the French foundation-model lab, scaled to millions of SDK downloads without a team hand-crafting each release - which is exactly the outcome Speakeasy is selling.
The Pivot Nobody Was Forced Into
Then the ground shifted, in the way it tends to. AI agents arrived, and with them the Model Context Protocol - MCP - a standard for letting language models call external tools. And an API, if you squint, is just a pile of tools an agent might want to use. Most infrastructure companies waited to see whether MCP was real. Speakeasy, which already turns specs into usable interfaces, moved. In 2025 it launched Gram, a platform for building and hosting MCP servers directly from your existing API, and then the Speakeasy MCP Gateway, an enterprise control plane pitched as "one entry point between every AI agent and every MCP server," complete with the security, governance, and observability that enterprises demand before they let a chatbot touch production.
There is a tidy logic here. The plumbing that REST APIs needed - authentication, versioning, observability, a way to expose capabilities safely - is exactly the plumbing AI agents now need. Speakeasy spent three years getting good at compiling one source of truth into many interfaces. Pointing that same machinery at agents is less a pivot than an extension. History, as they say, rhymes.
The Competition, and the Moat
Speakeasy is not alone. It competes with the venerable open-source OpenAPI Generator and Swagger Codegen - free, ubiquitous, and famous for output that reads like it was translated by a committee - as well as newer commercial players like Stainless, Fern, and liblab. The whole category is a bet that companies will pay to not think about SDKs. The differentiator Speakeasy keeps returning to is quality and maintenance: not just generating an SDK once, but keeping ten of them perpetually in sync with a spec that changes weekly, and doing it well enough that a developer can't tell a human didn't write it.
Whether that moat holds is the open question of the next few years, and it is complicated by the fact that the tool doing the generating is increasingly AI, which is available to everyone. But there is something durable about a company that owns the boring, universal middle of a workflow. The best developer infrastructure disappears into the pipeline; you stop thinking about it, which is the highest compliment a tool can earn. Speakeasy is trying to be the thing you forget is running - right up until the moment your API changes and, somewhere, ten SDKs quietly update themselves.
For a company selling a chore, that is a genuinely interesting place to stand.