Grafana open-sources an AI SDK for Go, wired to @ai-sdk/react
Go backends get streaming, tool calls and structured output that drop straight into React's useChat — plus smevals, a one-command uvx eval harness.
Copy markdown
Go finally speaks fluent ai-sdk
The SDK emits the @ai-sdk/react wire format, so a Go server now drives useChat, useCompletion and useObject on the frontend with zero glue code. Day-one providers: Anthropic (direct or via Vertex), OpenAI's Responses API, Amazon Bedrock, and any OpenAI-compatible endpoint.
Not a wrapper: agent loops and tool gates
Beyond chat it ships agent loops, structured-output validation and tool-approval workflows, plus observability and security docs aimed squarely at production. Runnable examples (agent-chat, structured-extraction) take you from install to a full-stack streaming app in three steps.
Elsewhere: eval a model over lunch
Simon Willison and Prime Radiant shipped smevals, a tiny framework that runs, grades and emits a static-HTML report for a model, prompt or harness — installable in one uvx command, no infra. Use it to sanity-check a cheaper model before you swap it into your stack.