Add parallel_tool_calls compatibility mapping

Preserve Chat Completions parallel_tool_calls when converting requests to the Responses API, and map Responses parallel_tool_calls back when falling back to Chat Completions upstreams.

Cover both true and explicit false values so clients can disable parallel tool calls without the field being dropped by omitempty.

backend/internal/pkg/apicompat/chatcompletions_responses_bridge.go

@@ -28,6 +28,7 @@ func ResponsesToChatCompletionsRequest(req *ResponsesRequest) (*ChatCompletionsR
 		TopP:                req.TopP,
 		Stream:              req.Stream,
 		ServiceTier:         req.ServiceTier,
+		ParallelToolCalls:   req.ParallelToolCalls,
 	}
 	if req.Reasoning != nil {
 		out.ReasoningEffort = req.Reasoning.Effort

backend/internal/pkg/apicompat/chatcompletions_responses_bridge_test.go

@@ -127,6 +127,26 @@ func TestResponsesToChatCompletionsRequest_TextFormatJsonSchema(t *testing.T) {
 	}`, string(out.ResponseFormat))
 }
 
+func TestResponsesToChatCompletionsRequest_ParallelToolCalls(t *testing.T) {
+	parallel := false
+	req := &ResponsesRequest{
+		Model: "gpt-4o",
+		Input: json.RawMessage(`[
+			{"role":"user","content":"Use tools"}
+		]`),
+		ParallelToolCalls: &parallel,
+	}
+
+	out, err := ResponsesToChatCompletionsRequest(req)
+	require.NoError(t, err)
+	require.NotNil(t, out.ParallelToolCalls)
+	assert.False(t, *out.ParallelToolCalls)
+
+	payload, err := json.Marshal(out)
+	require.NoError(t, err)
+	assert.Contains(t, string(payload), `"parallel_tool_calls":false`)
+}
+
 func chatMessageRoles(messages []ChatMessage) []string {
 	roles := make([]string, 0, len(messages))
 	for _, message := range messages {

backend/internal/pkg/apicompat/chatcompletions_responses_test.go

@@ -531,6 +531,25 @@ func TestChatCompletionsToResponses_ServiceTier(t *testing.T) {
 	assert.Equal(t, "flex", resp.ServiceTier)
 }
 
+func TestChatCompletionsToResponses_ParallelToolCalls(t *testing.T) {
+	for _, value := range []bool{false, true} {
+		req := &ChatCompletionsRequest{
+			Model:             "gpt-4o",
+			ParallelToolCalls: &value,
+			Messages:          []ChatMessage{{Role: "user", Content: json.RawMessage(`"Hi"`)}},
+		}
+
+		resp, err := ChatCompletionsToResponses(req)
+		require.NoError(t, err)
+		require.NotNil(t, resp.ParallelToolCalls)
+		assert.Equal(t, value, *resp.ParallelToolCalls)
+
+		payload, err := json.Marshal(resp)
+		require.NoError(t, err)
+		assert.Contains(t, string(payload), `"parallel_tool_calls":`+string(mustMarshalJSON(t, value)))
+	}
+}
+
 // ---------------------------------------------------------------------------
 // temperature / top_p stripping for reasoning models
 // ---------------------------------------------------------------------------

backend/internal/pkg/apicompat/chatcompletions_to_responses.go

@@ -27,12 +27,13 @@ func ChatCompletionsToResponses(req *ChatCompletionsRequest) (*ResponsesRequest,
 	}
 
 	out := &ResponsesRequest{
-		Model:        req.Model,
-		Instructions: req.Instructions,
-		Input:        inputJSON,
-		Stream:       true, // upstream always streams
-		Include:      []string{"reasoning.encrypted_content"},
-		ServiceTier:  req.ServiceTier,
+		Model:             req.Model,
+		Instructions:      req.Instructions,
+		Input:             inputJSON,
+		Stream:            true, // upstream always streams
+		Include:           []string{"reasoning.encrypted_content"},
+		ServiceTier:       req.ServiceTier,
+		ParallelToolCalls: req.ParallelToolCalls,
 	}
 
 	// Reasoning models (gpt-5.x) do not accept sampling parameters.

backend/internal/pkg/apicompat/types.go

@@ -435,6 +435,7 @@ type ChatCompletionsRequest struct {
 	Stream              bool               `json:"stream,omitempty"`
 	StreamOptions       *ChatStreamOptions `json:"stream_options,omitempty"`
 	Tools               []ChatTool         `json:"tools,omitempty"`
+	ParallelToolCalls   *bool              `json:"parallel_tool_calls,omitempty"`
 	ToolChoice          json.RawMessage    `json:"tool_choice,omitempty"`
 	ReasoningEffort     string             `json:"reasoning_effort,omitempty"` // "low" | "medium" | "high" | "xhigh"
 	ServiceTier         string             `json:"service_tier,omitempty"`