Reasoning Signature Part

[transparent163]

chat interface returns:

j:{"signature":"Reasoning"}
j:{"signature":" process c"}
j:{"signature":"ompleted by"}
j:{"signature":" Model-XYZ "}
j:{"signature":"v2.1 with "}
j:{"signature":"confiden"}
j:{"signature":"ce score 0."}
j:{"signature":"92 and veri"}
j:{"signature":"fication st"}
j:{"signature":"atus: pass"}
j:{"signature":"ed"}

But there is no push information in messages, what kind of data stream should I return?

[lgrammel]

This feature is very much Anthropic-specific, how are you using it?

[transparent163]

I am simulating an inference with a signature. I just made a change to return an inference content first and then the corresponding signature, but there is no signature field in the message.

go func() {
	defer close(event)
	textContent := "This is an AI-generated response with reasoning."
	event <- FormatDataContent(textContent, TextPart)
	time.Sleep(time.Duration(rand.Intn(400)+100) * time.Millisecond)
	reasoningContent := "Analyzing the given problem, considering possible solutions..."
	event <- FormatDataContent(reasoningContent, ReasoningPart)
	time.Sleep(time.Duration(rand.Intn(400)+100) * time.Millisecond)
	signatureContent := `{"signature": "abc123xyz", "model": "gpt-4-turbo", "confidence": 0.98, "verified": true}`
	event <- FormatDataContent(signatureContent, ReasoningSignaturePart)
}()

func FormatDataContent(data string, dataType StreamPartType) string {
	switch dataType {
	case TextPart:
		return fmt.Sprintf("0:\"%s\"\n", data)
	case ReasoningPart:
		return fmt.Sprintf("g:\"%s\"\n", data)
	case RedactedReasoningPart:
		return fmt.Sprintf("i:{\"data\":\"%s\"}\n", data)
	case ReasoningSignaturePart:
		return fmt.Sprintf("j:%s\n", data)

[github-actions[bot]]

This discussion was automatically locked because it has not been updated in over 30 days. If you still have questions about this topic, please ask us at community.vercel.com/ai-sdk