The Go AI SDK has reached 1.0. You call a config by key, and the variation it serves picks the model, provider, prompt, and tools. The SDK runs the call through your handler and reports cost, latency, tokens, and errors back to AgentControl. It includes handlers for OpenAI, Anthropic, and Amazon Bedrock, plus a LangChain-style handler for any other provider, and it supports streaming responses, multi-agent graphs, and online judges.
Go teams can now run offline evaluations from code. A single call sends each row of a dataset through your own handler and scores the output with checks you write in code, LaunchDarkly judges, or both. Each run returns pass or fail, so a regression can fail your build, and the results appear on the Evaluations page. To learn more, read https://launchdarkly.com/docs/home/agentcontrol/evaluations-from-code.
The Go SDK also supports Agent Skills. Skills attached to a config variation are delivered to your agent at runtime, so you can change an agent's instructions without shipping code. To learn more, read https://launchdarkly.com/docs/home/agentcontrol/agent-skills.
To upgrade, install https://github.com/launchdarkly/go-ai-sdk 1.0.0 or later with Go 1.26 or later. To learn more, read https://launchdarkly.com/docs/sdk/ai/go.
For help or feedback, contact your LaunchDarkly account team or email product@launchdarkly.com.