diff --git a/src/main/services/ai/human/node/brainNode/index.ts b/src/main/services/ai/human/node/brainNode/index.ts
index b9afb8db..6d811aad 100644
--- a/src/main/services/ai/human/node/brainNode/index.ts
+++ b/src/main/services/ai/human/node/brainNode/index.ts
@@ -25,6 +25,7 @@ export const createBrainNode = (input: { ctx: AiRunContext; cache: HumanRuntimeC
       schema: humanThoughtDecisionSchema,
       schemaName: 'human_brain_decision',
       temperature: 0.1,
+      streamingMode: 'force_non_stream',
       contextBudgetSnapshot: state.execution.contextBudget
     })
     const inputTokens = result.usage?.inputTokens
diff --git a/src/main/services/ai/llmExecutor.ts b/src/main/services/ai/llmExecutor.ts
index 9738ca82..7aade1fc 100644
--- a/src/main/services/ai/llmExecutor.ts
+++ b/src/main/services/ai/llmExecutor.ts
@@ -113,6 +113,7 @@ export type AiLlmTextRequest = {
   system: string
   user: string
   temperature?: number
+  streamingMode?: AiLlmStreamingMode
   onTextDelta?: (text: string) => void
 }
 
@@ -124,6 +125,7 @@ export type AiLlmStructuredRequest<T extends Record<string, unknown>> = {
   user: string
   schema: z.ZodType<T>
   temperature?: number
+  streamingMode?: AiLlmStreamingMode
   contextBudgetSnapshot?: BrainContextBudgetSnapshot
 }
 
@@ -706,7 +708,7 @@ export const createAiLlmExecutor = (input: {
   checkCancelled: () => void
   driver?: AiLlmDriver
 }): AiLlmExecutor => {
-  const streamingMode = resolveStreamingMode(input.runtimeSettings)
+  const defaultStreamingMode = resolveStreamingMode(input.runtimeSettings)
 
   return {
     chatStructured: async <T extends Record<string, unknown>>(
@@ -716,6 +718,7 @@ export const createAiLlmExecutor = (input: {
       usage?: AiLlmUsage
     }> => {
       input.checkCancelled()
+      const streamingMode = request.streamingMode || defaultStreamingMode
       const cfg = await input.getConfig()
       const result = await runLlmCallWithActivity<Record<string, unknown>>({
         ctx: input.eventCtx,
@@ -777,6 +780,7 @@ export const createAiLlmExecutor = (input: {
 
     chatText: async (request: AiLlmTextRequest): Promise<string> => {
       input.checkCancelled()
+      const streamingMode = request.streamingMode || defaultStreamingMode
       const cfg = await input.getConfig()
       const result = await runLlmCallWithActivity<string>({
         ctx: input.eventCtx,
diff --git a/src/main/services/ai/types.ts b/src/main/services/ai/types.ts
index 9d637a67..3de7912e 100644
--- a/src/main/services/ai/types.ts
+++ b/src/main/services/ai/types.ts
@@ -16,6 +16,7 @@ import type {
   AiApprovalBatchDecision,
   AiApprovalBatchItem,
   AiLlmIoRecord,
+  AiLlmStreamingMode,
   AiLlmUsage,
   AiRuntimeSettings,
   AiRunEvent,
@@ -135,6 +136,7 @@ export type AiRunContext = {
     schema: z.ZodType<T>
     schemaName: string
     temperature?: number
+    streamingMode?: AiLlmStreamingMode
     contextBudgetSnapshot?: BrainContextBudgetSnapshot
   }) => Promise<{
     output: T
@@ -147,6 +149,7 @@ export type AiRunContext = {
     user: string
     schemaName: string
     temperature?: number
+    streamingMode?: AiLlmStreamingMode
     onTextDelta?: (text: string) => void
   }) => Promise<string>
   sendEvent: (event: AiRunEvent) => void