Google's Genkit Go SDK hit its first stable release June 18, committing to backward API compatibility and shipping Claude, Vertex AI and Ollama plugins.
Google declared Genkit’s Go SDK production-ready on June 18, committing to backward API compatibility across all future 1.x releases. Programs written against Genkit Go 1.x will compile unchanged as minor versions accumulate. That’s the guarantee Go developers need from any library they put in a production backend.
The announcement came via the Google Developers Blog (June 18, 2026).
What Stable Means Here #
Before this release, the Go SDK was explicitly experimental — API surfaces moved between minor versions, and production use carried risk. The 1.x stability promise changes that: the call signatures you write today will not break when Genkit Go 1.1 or 1.2 ships. That’s a different contract than “works for us internally,” and it’s what Go teams need before adopting a dependency in backend services.
The SDK is now at v1.10.0 (published June 26, 2026), reflecting active development since the 1.0 announcement.
What Ships in the Stable SDK #
The stable Go module includes production-ready support for:
Type-safe AI flows—genkit.DefineFlow
compiles the input/output contract into Go’s type system, so schema mismatches are caught at build time, not runtime.Unified model interface— one API surface across Google AI (Gemini), Vertex AI, Anthropic Claude and Ollama. Provider-switching is a plugin swap, not a code rewrite.Tool calling and RAG—ai.DefineTool
and retriever integrations (Pinecone, Weaviate, PostgreSQL pgvector) ship as first-class plugins, not experimental extras.Structured outputs— responses can be constrained to a Go struct viaai.WithOutputSchema
; the framework validates and retries on malformed JSON automatically.Streaming— durable streaming is incore/x/streaming
(marked experimental within the otherwise-stable module, per the release notes).Agentic workflows— multi-turn agents with session stores are inai/exp
; they follow the same experimental-within-stable pattern as JS’s early agent API.
The Anthropic plugin extends the SDK beyond Google’s own model stack. Ollama enables local-model development without network calls, useful for CI pipelines and air-gapped environments.
The genkit init:ai-tools #
Command
The release also ships a new CLI command: genkit init:ai-tools
. It generates a .genkit/
config directory that AI coding tools (Gemini CLI, Claude Code, Cursor) pick up automatically, giving them type-aware context about your Genkit flows and plugins without manual prompting. The integrations are opt-in and add no runtime dependency.
Where Go Fits in the Genkit Ecosystem #
Genkit JS is the more mature SDK (at v1.39, with production agents, Firestore session persistence, and Vercel AI SDK integration). Genkit Python remains in alpha. Go is now the second stable runtime. For Firebase backend teams running high-throughput Cloud Run services or gRPC APIs, where container size and concurrency count, the JS SDK doesn’t fit. This release gives those teams a production-grade option in their own language.
The practical split is: JS for Firebase AI Logic and web-adjacent backend code; Go for high-throughput Cloud Run services, gRPC APIs, and teams that have already standardized on Go for their Firebase backend.
Key Takeaways #
API stability is the headline: Genkit Go 1.x is a stable dependency contract, not just a milestone version.** Anthropic and Ollama**are production-supported plugins alongside Google AI and Vertex AI.wires your Genkit project into AI coding assistants.genkit init:ai-tools
Streaming and agents ship as experimental sub-packages within the stable module — available to use, but outside the compatibility guarantee until they graduate.- Python stays alpha; Go is now the second production-stable Genkit runtime.
Related #
— what Genkit JS 1.39 brings for server-side agent development.Genkit Retires the Chat Beta for Production Agents