Partnering with Google to open source the foundations of DX & AX
Sagar Batchu
September 17, 2026 · 4 min read
Today we’re announcing our partnership with Google to open source the Speakeasy client generator. At Speakeasy, we are building the AI control plane to govern how agents safely access enterprise data. That access starts with connection, and connection means APIs that are easy to integrate and maintain. That’s where we’ve been collaborating with the Google team.
Developers now get open source access to SDK generation in seven languages, MCP server generation, and CLI generation. Everything that is needed to maintain a best-in-class developer experience (DX) and agent experience (AX) for their services.
How the partnership started
The SDKs for the Interactions, Agents, and Webhooks APIs have always been generated from API descriptions. The Google team started in-house, before moving to a vendor solution in late 2025. In May 2026, just as the team was preparing for Google I/O and the general availability of the Interactions API, that vendor was acquired and announced its shutdown. The Google team suddenly needed a new path forward for its generation pipeline, and chose to delay the GA of the Interactions API rather than rush out broken SDKs.
We partnered with Google and worked in tight collaboration to move the SDKs onto Speakeasy generation, with minimal disruption for the developers who rely on them. That meant aligning type definitions across every target language, preserving the existing error hierarchies and streaming behavior, and integrating the generator directly into Google’s internal monorepo and build system.
Here is what calling the Interactions API looks like through the generated Python SDK today:
from google import genai
client = genai.Client()
interaction = client.interactions.create( model="gemini-3.8-flash", input="Say warningly: Moat is not a noun. It's a verb.", response_format={"type": "audio"}, generation_config={ "speech_config": [ {"voice": "Kore"} ] })From DX to AX
What began as a migration turned into a broader collaboration. As we worked through the SDK layer, a second question emerged. What do agents need from an API?
The first result of that new work is the Gemini Docs MCP server, built with Google on the same foundation. It gives AI coding agents direct access to Google’s maintained, markdown-first documentation. An agent writing code against the Gemini API learns from the current docs rather than whatever version was in its training data, and the server nudges agents toward using the SDKs instead of hallucinating their own integrations. The docs MCP server is also pre-bundled into Antigravity to improve use of the Gemini API.
Google is leading the charge on future-proofing its integration experience to make sure that agents have the same delightful experience as developers.
What’s included in the release
Having lived through a critical dependency disappearing, Google wanted to make sure the ecosystem didn’t suffer another rug pull in the future. API artifacts are critical dependencies, and they require continuity.
Our team wholeheartedly agreed and was eager to work with Google on open sourcing the generator. Here’s what’s included:
SDK generators
Seven languages: TypeScript, Python, Go, Java, C#, PHP, and Ruby.
CLI generator
Drive any API ad hoc from a terminal, whether the caller is a developer or an agent.
MCP server generator
Expose an API as tools that agents can call at runtime.
Docs MCP server
A hosted server that indexes any markdown repo and makes it searchable by an agent.
The entire suite is released under AGPL-3.0. Code generated with these tools belongs to you in perpetuity. Commercial licensing is still available directly from Speakeasy for teams that need it.
Both developers and agents deserve first-class support, and both should be able to rely on the tooling being there now and forever.
What’s next for Speakeasy
We are building the AI control plane, the layer that governs how agents connect to real systems, with the identity, policy, and observability that enterprises require once agents start making calls that matter.
The generators being open sourced today make it easy for any company to make its data agent-accessible. The control plane makes sure that access is secure and governed.
Our ongoing commitment
Backing this release is one step toward making sure that critical tooling stays dependable regardless of who owns it.
Our thanks to the Google team for moving quickly, and for helping us put this work in the open. We look forward to continuing to innovate together to give Gemini a great developer and agent experience, and to making those foundations available to developers everywhere.
Get started with the open source repository.
Last updated on