translator: accept thinking_blocks signatures and fall back to skip_thought_signature_validator for Gemini function calls

internal/apischema/openai/openai.go

@@ -641,6 +641,10 @@ type ChatCompletionAssistantMessageParam struct {
 	Refusal string `json:"refusal,omitempty"`
 	// The tool calls generated by the model, such as function calls.
 	ToolCalls []ChatCompletionMessageToolCallParam `json:"tool_calls,omitempty"`
+	// ThinkingBlocks holds structured thinking metadata (signatures, redacted content) echoed back
+	// by clients from a previous response's thinking_blocks. Non-standard extension, see
+	// https://docs.litellm.ai/docs/reasoning_content for the convention.
+	ThinkingBlocks []ThinkingBlock `json:"thinking_blocks,omitempty"`
 }
 
 // ChatCompletionMessageToolCallType The type of the tool. Currently, only `function` is supported.

internal/translator/gemini_helper.go

@@ -33,6 +33,11 @@ const (
 	gcpMethodRawPredict            = "rawPredict"
 )
 
+// dummyThoughtSignature is Google's documented compatibility escape for clients that cannot echo
+// back a thought_signature on function-call parts in multi-turn requests, see
+// https://ai.google.dev/gemini-api/docs/thought-signatures.
+var dummyThoughtSignature = []byte("skip_thought_signature_validator")
+
 // geminiResponseMode represents the type of response mode for Gemini requests
 type geminiResponseMode string
 
@@ -275,6 +280,21 @@ func assistantMsgToGeminiParts(msg *openai.ChatCompletionAssistantMessageParam)
 		}
 	}
 
+	// Fall back to the first non-empty signature echoed back via thinking_blocks (see
+	// https://docs.litellm.ai/docs/reasoning_content) when no content-part signature was found.
+	if thoughtSignature == nil {
+		for _, block := range msg.ThinkingBlocks {
+			if block.Signature != "" {
+				sigBytes, err := base64.StdEncoding.DecodeString(block.Signature)
+				if err != nil {
+					return nil, nil, fmt.Errorf("failed to decode thought signature: %w", err)
+				}
+				thoughtSignature = sigBytes
+				break // Only use first signature.
+			}
+		}
+	}
+
 	// Handle tool calls in the assistant message.
 	knownToolCalls := make(map[string]string)
 	for i, toolCall := range msg.ToolCalls {
@@ -290,8 +310,16 @@ func assistantMsgToGeminiParts(msg *openai.ChatCompletionAssistantMessageParam)
 		funcCallPart := genai.NewPartFromFunctionCall(toolCall.Function.Name, parsedArgs)
 
 		// According to https://ai.google.dev/gemini-api/docs/thought-signatures, if the model generates parallel function calls in a response, the thought_signature is attached only to the first functionCall part. Subsequent  functionCall parts in the same response will not contain a signature.
-		if i == 0 && thoughtSignature != nil {
-			funcCallPart.ThoughtSignature = thoughtSignature
+		if i == 0 {
+			if thoughtSignature != nil {
+				funcCallPart.ThoughtSignature = thoughtSignature
+			} else {
+				// No signature was echoed back by the client at all (e.g. an OpenAI-schema client
+				// that doesn't round-trip thinking_blocks). Gemini 3.x rejects function calls in
+				// multi-turn requests that are missing a thought_signature, so fall back to
+				// Google's documented compatibility escape.
+				funcCallPart.ThoughtSignature = dummyThoughtSignature
+			}
 		}
 
 		parts = append(parts, funcCallPart)

internal/translator/gemini_helper_test.go

@@ -94,6 +94,7 @@ func TestOpenAIMessagesToGeminiContents(t *testing.T) {
 									"param1": "value1",
 								},
 							},
+							ThoughtSignature: dummyThoughtSignature,
 						},
 						{Text: "This is a assistant message"},
 					},
@@ -249,6 +250,7 @@ func TestAssistantMsgToGeminiParts(t *testing.T) {
 						Args: map[string]any{"location": "New York", "unit": "celsius"},
 						Name: "get_weather",
 					},
+					ThoughtSignature: dummyThoughtSignature,
 				},
 			},
 			expectedToolCalls: map[string]string{
@@ -282,10 +284,14 @@ func TestAssistantMsgToGeminiParts(t *testing.T) {
 				},
 			},
 			expectedParts: []*genai.Part{
-				genai.NewPartFromFunctionCall("get_weather", map[string]any{
-					"location": "New York",
-					"unit":     "celsius",
-				}),
+				func() *genai.Part {
+					p := genai.NewPartFromFunctionCall("get_weather", map[string]any{
+						"location": "New York",
+						"unit":     "celsius",
+					})
+					p.ThoughtSignature = dummyThoughtSignature
+					return p
+				}(),
 				genai.NewPartFromFunctionCall("get_time", map[string]any{
 					"timezone": "EST",
 				}),
@@ -428,9 +434,13 @@ func TestAssistantMsgToGeminiParts(t *testing.T) {
 				},
 			},
 			expectedParts: []*genai.Part{
-				genai.NewPartFromFunctionCall("get_weather", map[string]any{
-					"location": "San Francisco",
-				}),
+				func() *genai.Part {
+					p := genai.NewPartFromFunctionCall("get_weather", map[string]any{
+						"location": "San Francisco",
+					})
+					p.ThoughtSignature = dummyThoughtSignature
+					return p
+				}(),
 				{
 					Text:    "I need to call a function to get the weather",
 					Thought: true,
@@ -520,6 +530,159 @@ func TestAssistantMsgToGeminiParts(t *testing.T) {
 				"call_time":    "get_time",
 			},
 		},
+		{
+			name: "thinking_blocks signature with tool calls - signature on first tool call",
+			msg: openai.ChatCompletionAssistantMessageParam{
+				Role: openai.ChatMessageRoleAssistant,
+				ThinkingBlocks: []openai.ThinkingBlock{
+					{Type: "thinking", Signature: "dGVzdHNpZ25hdHVyZQ=="}, // "testsignature" in base64
+				},
+				ToolCalls: []openai.ChatCompletionMessageToolCallParam{
+					{
+						ID: ptr.To("call_weather"),
+						Function: openai.ChatCompletionMessageToolCallFunctionParam{
+							Name:      "get_weather",
+							Arguments: `{"location":"San Francisco"}`,
+						},
+						Type: openai.ChatCompletionMessageToolCallTypeFunction,
+					},
+					{
+						ID: ptr.To("call_time"),
+						Function: openai.ChatCompletionMessageToolCallFunctionParam{
+							Name:      "get_time",
+							Arguments: `{"timezone":"PST"}`,
+						},
+						Type: openai.ChatCompletionMessageToolCallTypeFunction,
+					},
+				},
+			},
+			expectedParts: []*genai.Part{
+				{
+					FunctionCall: &genai.FunctionCall{
+						Name: "get_weather",
+						Args: map[string]any{"location": "San Francisco"},
+					},
+					ThoughtSignature: []byte("testsignature"),
+				},
+				{
+					FunctionCall: &genai.FunctionCall{
+						Name: "get_time",
+						Args: map[string]any{"timezone": "PST"},
+					},
+				},
+			},
+			expectedToolCalls: map[string]string{
+				"call_weather": "get_weather",
+				"call_time":    "get_time",
+			},
+		},
+		{
+			name: "tool calls with no signature anywhere fall back to dummy signature on first call only",
+			msg: openai.ChatCompletionAssistantMessageParam{
+				Role: openai.ChatMessageRoleAssistant,
+				ToolCalls: []openai.ChatCompletionMessageToolCallParam{
+					{
+						ID: ptr.To("call_weather"),
+						Function: openai.ChatCompletionMessageToolCallFunctionParam{
+							Name:      "get_weather",
+							Arguments: `{"location":"San Francisco"}`,
+						},
+						Type: openai.ChatCompletionMessageToolCallTypeFunction,
+					},
+					{
+						ID: ptr.To("call_time"),
+						Function: openai.ChatCompletionMessageToolCallFunctionParam{
+							Name:      "get_time",
+							Arguments: `{"timezone":"PST"}`,
+						},
+						Type: openai.ChatCompletionMessageToolCallTypeFunction,
+					},
+				},
+			},
+			expectedParts: []*genai.Part{
+				{
+					FunctionCall: &genai.FunctionCall{
+						Name: "get_weather",
+						Args: map[string]any{"location": "San Francisco"},
+					},
+					ThoughtSignature: dummyThoughtSignature,
+				},
+				{
+					FunctionCall: &genai.FunctionCall{
+						Name: "get_time",
+						Args: map[string]any{"timezone": "PST"},
+					},
+				},
+			},
+			expectedToolCalls: map[string]string{
+				"call_weather": "get_weather",
+				"call_time":    "get_time",
+			},
+		},
+		{
+			name: "content thinking-part signature wins over differing thinking_blocks signature",
+			msg: openai.ChatCompletionAssistantMessageParam{
+				Content: openai.StringOrAssistantRoleContentUnion{
+					Value: []openai.ChatCompletionAssistantMessageParamContent{
+						{
+							Type:      openai.ChatCompletionAssistantMessageParamContentTypeThinking,
+							Text:      ptr.To("I need to call a function to get the weather"),
+							Signature: ptr.To("dGVzdHNpZ25hdHVyZQ=="), // "testsignature" in base64
+						},
+					},
+				},
+				Role: openai.ChatMessageRoleAssistant,
+				ThinkingBlocks: []openai.ThinkingBlock{
+					{Type: "thinking", Signature: "b3RoZXJzaWduYXR1cmU="}, // "othersignature" in base64
+				},
+				ToolCalls: []openai.ChatCompletionMessageToolCallParam{
+					{
+						ID: ptr.To("call_weather"),
+						Function: openai.ChatCompletionMessageToolCallFunctionParam{
+							Name:      "get_weather",
+							Arguments: `{"location":"San Francisco"}`,
+						},
+						Type: openai.ChatCompletionMessageToolCallTypeFunction,
+					},
+				},
+			},
+			expectedParts: []*genai.Part{
+				{
+					FunctionCall: &genai.FunctionCall{
+						Name: "get_weather",
+						Args: map[string]any{"location": "San Francisco"},
+					},
+					ThoughtSignature: []byte("testsignature"),
+				},
+				{
+					Text:    "I need to call a function to get the weather",
+					Thought: true,
+				},
+			},
+			expectedToolCalls: map[string]string{
+				"call_weather": "get_weather",
+			},
+		},
+		{
+			name: "invalid base64 in thinking_blocks signature",
+			msg: openai.ChatCompletionAssistantMessageParam{
+				Role: openai.ChatMessageRoleAssistant,
+				ThinkingBlocks: []openai.ThinkingBlock{
+					{Type: "thinking", Signature: "not-valid-base64!!!"},
+				},
+				ToolCalls: []openai.ChatCompletionMessageToolCallParam{
+					{
+						ID: ptr.To("call_weather"),
+						Function: openai.ChatCompletionMessageToolCallFunctionParam{
+							Name:      "get_weather",
+							Arguments: `{"location":"San Francisco"}`,
+						},
+						Type: openai.ChatCompletionMessageToolCallTypeFunction,
+					},
+				},
+			},
+			expectedErrorMsg: "failed to decode thought signature",
+		},
 		{
 			name: "thinking content with invalid base64 signature",
 			msg: openai.ChatCompletionAssistantMessageParam{