feat: add Gemini API compatibility, refactor stream rendering, and enhance tool call handling and configuration options

internal/format/openai/render_responses.go

@@ -21,12 +21,6 @@ func BuildResponseObject(responseID, model, finalPrompt, finalThinking, finalTex
 	output := make([]any, 0, 2)
 	if len(detected) > 0 {
 		exposedOutputText = ""
-		if strings.TrimSpace(finalThinking) != "" {
-			output = append(output, map[string]any{
-				"type": "reasoning",
-				"text": finalThinking,
-			})
-		}
 		output = append(output, toResponsesFunctionCallItems(detected)...)
 	} else {
 		content := make([]any, 0, 2)
@@ -52,6 +46,21 @@ func BuildResponseObject(responseID, model, finalPrompt, finalThinking, finalTex
 			"content": content,
 		})
 	}
+	return BuildResponseObjectFromItems(
+		responseID,
+		model,
+		finalPrompt,
+		finalThinking,
+		finalText,
+		output,
+		exposedOutputText,
+	)
+}
+
+func BuildResponseObjectFromItems(responseID, model, finalPrompt, finalThinking, finalText string, output []any, outputText string) map[string]any {
+	if output == nil {
+		output = []any{}
+	}
 	return map[string]any{
 		"id":          responseID,
 		"type":        "response",
@@ -60,7 +69,7 @@ func BuildResponseObject(responseID, model, finalPrompt, finalThinking, finalTex
 		"status":      "completed",
 		"model":       model,
 		"output":      output,
-		"output_text": exposedOutputText,
+		"output_text": outputText,
 		"usage":       BuildResponsesUsage(finalPrompt, finalThinking, finalText),
 	}
 }

internal/format/openai/render_stream_events.go

@@ -59,12 +59,15 @@ func BuildResponsesContentPartDonePayload(responseID, itemID string, outputIndex
 	}
 }
 
-func BuildResponsesTextDeltaPayload(responseID, delta string) map[string]any {
+func BuildResponsesTextDeltaPayload(responseID, itemID string, outputIndex, contentIndex int, delta string) map[string]any {
 	return map[string]any{
-		"type":        "response.output_text.delta",
-		"id":          responseID,
-		"response_id": responseID,
-		"delta":       delta,
+		"type":          "response.output_text.delta",
+		"id":            responseID,
+		"response_id":   responseID,
+		"item_id":       itemID,
+		"output_index":  outputIndex,
+		"content_index": contentIndex,
+		"delta":         delta,
 	}
 }
 

internal/format/openai/render_test.go

@@ -138,15 +138,11 @@ func TestBuildResponseObjectDetectsToolCallFromThinkingChannel(t *testing.T) {
 	)
 
 	output, _ := obj["output"].([]any)
-	if len(output) != 2 {
-		t.Fatalf("expected reasoning + function_call outputs, got %#v", obj["output"])
+	if len(output) != 1 {
+		t.Fatalf("expected function_call output only, got %#v", obj["output"])
 	}
 	first, _ := output[0].(map[string]any)
-	if first["type"] != "reasoning" {
-		t.Fatalf("expected first output reasoning, got %#v", first["type"])
-	}
-	second, _ := output[1].(map[string]any)
-	if second["type"] != "function_call" {
-		t.Fatalf("expected second output function_call, got %#v", second["type"])
+	if first["type"] != "function_call" {
+		t.Fatalf("expected output function_call, got %#v", first["type"])
 	}
 }