fix(ai): honor Responses parallel tool control (#44744)

packages/ai/src/protocols/open-responses.ts

@@ -666,6 +666,7 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
 const lowerOptions = (request: LLMRequest) => {
   const options = OpenResponsesOptions.resolve(request)
   const cacheKey = ProviderShared.clampPromptCacheKey(request.promptCacheKey)
+  const parallelToolCalls = resolveParallelToolCalls(request)
   return {
     ...(options.instructions ? { instructions: options.instructions } : {}),
     ...(options.store !== undefined ? { store: options.store } : {}),
@@ -683,11 +684,18 @@ const lowerOptions = (request: LLMRequest) => {
     ...(options.textVerbosity ? { text: { verbosity: options.textVerbosity } } : {}),
     ...(options.serviceTier ? { service_tier: options.serviceTier } : {}),
     ...(options.maxToolCalls !== undefined ? { max_tool_calls: options.maxToolCalls } : {}),
-    ...(options.parallelToolCalls !== undefined ? { parallel_tool_calls: options.parallelToolCalls } : {}),
+    ...(parallelToolCalls !== undefined ? { parallel_tool_calls: parallelToolCalls } : {}),
     ...(options.truncation ? { truncation: options.truncation } : {}),
   }
 }
 
+export const resolveParallelToolCalls = (request: LLMRequest) => {
+  const configured = OpenResponsesOptions.resolve(request).parallelToolCalls
+  if (configured !== undefined) return configured
+  const disabled = request.toolChoice?.disableParallelToolUse
+  return disabled === undefined ? undefined : !disabled
+}
+
 const allowedToolChoice = (request: LLMRequest) => {
   const allowed = OpenResponsesOptions.resolve(request).allowedTools
   if (!allowed) return undefined

packages/ai/src/protocols/openai-responses.ts

@@ -111,8 +111,10 @@ const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request:
     extension,
   )
   const toolSchemaCompatibility = request.model.compatibility?.toolSchema
+  const parallelToolCalls = OpenResponses.resolveParallelToolCalls(request)
   return {
     ...body,
+    ...(parallelToolCalls === undefined ? {} : { parallel_tool_calls: parallelToolCalls }),
     tools:
       request.tools.length === 0
         ? undefined

packages/ai/test/provider/openai-compatible-responses.test.ts

@@ -132,6 +132,40 @@ describe("Open Responses-compatible route", () => {
     }),
   )
 
+  it.effect("lowers canonical parallel tool control", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+      }).model("example-model")
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          prompt: "Read the file.",
+          tools: [
+            ToolDefinition.make({
+              name: "read",
+              description: "Read a file.",
+              inputSchema: { type: "object" },
+            }),
+          ],
+          toolChoice: { type: "auto", disableParallelToolUse: true },
+        }),
+      )
+
+      expect(prepared.body.parallel_tool_calls).toBe(false)
+      expect(prepared.body.tools).toEqual([
+        {
+          type: "function",
+          name: "read",
+          description: "Read a file.",
+          parameters: { type: "object" },
+          strict: false,
+        },
+      ])
+    }),
+  )
+
   it.effect("keeps foreign item id grammars but drops malformed ids", () =>
     Effect.gen(function* () {
       const model = configure({

packages/ai/test/provider/openai-responses.test.ts

@@ -258,6 +258,31 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
+  it.effect("maps the canonical parallel tool setting with provider-option precedence", () =>
+    Effect.gen(function* () {
+      const disabled = yield* compileRequest(
+        LLMRequest.update(request, {
+          toolChoice: { type: "auto", disableParallelToolUse: true },
+        }),
+      )
+      const enabled = yield* compileRequest(
+        LLMRequest.update(request, {
+          toolChoice: { type: "auto", disableParallelToolUse: false },
+        }),
+      )
+      const overridden = yield* compileRequest(
+        LLMRequest.update(request, {
+          toolChoice: { type: "auto", disableParallelToolUse: true },
+          providerOptions: { parallelToolCalls: true },
+        }),
+      )
+
+      expect(disabled.body.parallel_tool_calls).toBe(false)
+      expect(enabled.body.parallel_tool_calls).toBe(true)
+      expect(overridden.body.parallel_tool_calls).toBe(true)
+    }),
+  )
+
   it.effect("lowers chronological system updates to developer messages in order", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(