Go LLM SDK for streaming, tool-calling AI backends (plus frontend React lib)

Who Should Read

A backend developer building an AI chat backend in Go who wants to connect a React useChat frontend without a protocol adapter

Core Mechanics

• Go talks directly to Vercel AI SDK frontends
  • It is designed to be wire-compatible with the Vercel AI SDK's TypeScript frontend hooks.
  • A Go endpoint streams Server-Sent Events (SSE) directly to hooks such as useChat.
• No protocol adapter when swapping backends
  • You can reuse an existing AI SDK React frontend.
  • You can also replace a TypeScript backend with Go without an adapter.
• One API for calls, streaming, tools, and agents
  • StreamText and GenerateText stream a response or return the full result, with retries and multi-step tool execution.
  • It also generates schema-validated structured output (objects, arrays, choices).
• Plain Go functions as tools, with approval
  • A model can call plain Go functions.
  • Consequential actions can require approval.
• Providers and production controls included

Community Reactions

• A commenter asked what the SDK is for
  • They asked why use it instead of calling the OpenAI API directly.
  • A reply explained it is a Go backend wire-compatible with the Vercel AI UI SDK.
• Resumable streaming is missing, one commenter said
  • They said it misses resumable streaming after a dropped SSE connection, and multi-device support.
  • Their view is that such SDKs are easy to drop in but hard to build resilient systems on.
• Requests to compare with similar SDKs
  • Commenters asked how it compares to genkit and Google's ADK.
  • One commenter said they have been building a Go project called gollem for five months.
• Skepticism about AI-generated code
  • One commenter asked whether a human designed it or it is AI slop.
  • They said they are much more conservative about adding packages to production and would review it later.

How to Apply

• Run the quick-start example locally
  • Create a Go module and go get the core and anthropic provider packages.
  • Set ANTHROPIC_API_KEY and run the GenerateText example with go run.
• Point your useChat frontend at a Go endpoint
  • Change the URL your AI SDK React transport points to so it hits the Go server.
  • On the server, write the StreamText result to SSE with WriteUIMessageStream.
• Test reconnect behavior before adopting
  • Cut the network mid-stream and observe what happens on reconnect.
  • If you need recovery, decide on a separate layer first.
• Put approval gates on risky tools
  • Mark high-impact Go functions like payments or deletes as requiring approval.
  • Expose read-only functions as tools without approval.

Try It

# Run in a terminal: quick start (install and example as in the README)

mkdir ai-sdk-quickstart && cd ai-sdk-quickstart
go mod init example.com/ai-sdk-quickstart
go get github.com/grafana/ai-sdk
go get github.com/grafana/ai-sdk/providers/anthropic

# Save as main.go
package main

import (
	"context"
	"fmt"
	"log"
	"os"

	aisdk "github.com/grafana/ai-sdk"
	"github.com/grafana/ai-sdk/provider"
	"github.com/grafana/ai-sdk/providers/anthropic"
)

func main() {
	apiKey := os.Getenv("ANTHROPIC_API_KEY")
	if apiKey == "" {
		log.Fatal("ANTHROPIC_API_KEY is required")
	}
	model := anthropic.New(apiKey, "claude-sonnet-5")
	result, err := aisdk.GenerateText(context.Background(), model,
		aisdk.WithModelMessages(provider.UserText("Explain goroutines in one sentence.")),
	)
	if err != nil {
		log.Fatal(err)
	}
	fmt.Println(result.Text)
}

# Run
ANTHROPIC_API_KEY=sk-... go run .

# Next: follow Build a full-stack chat in the docs to connect StreamText to React useChat

Terminology

Related Papers

• Same Simulation, Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy

  Saving long tool output to files cut GPT-5.1 accuracy by 11.3 points; naming the authoritative file restored it.

• Fake News, Real Crashes: Information Injection Attacks on LLM Financial Trading Agents

  Fake headlines hijacked LLM trading agents 62% of the time versus 14% for prompt injection.

• Harness Engineering 101: The LLM is the brain, the harness is the body

  An agent is a loop managing a JSON array; 16 chapters build one from scratch

• Harness2 by Google Research – Recursive Agent Harnessing

  Recursively refining the harness with fixed model weights raised benchmark scores by up to 11.8 points.

• Ousted Savannah police officer gave ICE agent husband access to Flock system

  A police officer shared her Flock login with her ICE-agent husband; an audit found 983 searches and she was fired

• Tmux and Claude Code: Sessions, Send-Keys, and the Ceiling

  Running Claude Code agents in tmux works well, but limits appear around four agents.

Related Resources

• https://github.com/grafana/ai-sdk
• https://github.com/genkit-ai/genkit
• https://github.com/fugue-labs/gollem