backend-only.md

Generate text from Go

Starting from the model created in Installation, use GenerateText when application code needs one complete result and StreamText when a caller should receive text as it arrives. Both accept the same instructions, messages, tools, provider options, and reliability settings.

A system instruction defines how the model should behave. A user message carries the request to answer.

Generate a complete response

GenerateText blocks until the operation finishes:

result, err := aisdk.GenerateText(ctx, model,
	aisdk.WithSystem("You write concise incident summaries."),
	aisdk.WithModelMessages(
		provider.UserText("Summarize the payment API incident."),
	),
)
if err != nil {
	return err
}

fmt.Println(result.Text)
fmt.Println(result.TotalUsage)

Use this path for background jobs, API endpoints that return one JSON response, classification pipelines, and other work that does not need incremental output. The result also exposes tool calls, sources, files, warnings, finish reason, and per-step information.

Stream text as it arrives

StreamText returns immediately. Consume its stream exactly once:

result := aisdk.StreamText(ctx, model,
	aisdk.WithModelMessages(provider.UserText("Explain this stack trace.")),
)

for part := range result.FullStream() {
	switch part := part.(type) {
	case aisdk.StreamTextDelta:
		fmt.Print(part.Text)
	case aisdk.StreamError:
		log.Printf("model stream: %v", part.Error)
	}
}
if err := result.Err(); err != nil {
	return err
}

Reading FullStream, converting with ToUIMessageStream, or passing the result to an HTTP writer all consume the same underlying stream. Choose one consumer. Blocking accessors such as Text() and TotalUsage() are populated after the stream finishes.

The loop above is the complete terminal-consumption pattern. For larger application outcomes, continue to the runnable examples.

Choose the next capability

• Need typed JSON? Use structured output.
• Need the model to call application code? Add tools.
• Need several model/tool turns? Configure an agent loop.
• Need a React client? Build a full-stack chat.

For exact result fields and options, see GenerateText and StreamText.

---

← Installation · Docs index · Full-stack chat →