diff --git a/src/main/apiServer/middleware/gateway.ts b/src/main/apiServer/middleware/gateway.ts
index 4fa4f5f77f..2de7a6a726 100644
--- a/src/main/apiServer/middleware/gateway.ts
+++ b/src/main/apiServer/middleware/gateway.ts
@@ -7,13 +7,210 @@
  * - Model injection for simplified external app integration
  */
 
+import { buildFunctionCallToolName } from '@main/utils/mcp'
+import type { MCPTool } from '@types'
 import type { NextFunction, Request, Response } from 'express'
 
 import { loggerService } from '../../services/LoggerService'
+import { reduxService } from '../../services/ReduxService'
 import { config } from '../config'
+import { mcpApiService } from '../services/mcp'
+import { getMCPServersFromRedux } from '../utils/mcp'
 
 const logger = loggerService.withContext('GatewayMiddleware')
 
+type AssistantConfig = {
+  id: string
+  name: string
+  prompt?: string
+  model?: { id: string; provider: string }
+  defaultModel?: { id: string; provider: string }
+  settings?: {
+    streamOutput?: boolean
+    enableTemperature?: boolean
+    temperature?: number
+    enableTopP?: boolean
+    topP?: number
+    enableMaxTokens?: boolean
+    maxTokens?: number
+  }
+  mcpServers?: Array<{ id: string }>
+  allowed_tools?: string[]
+}
+
+type ToolDefinition = {
+  name: string
+  description?: string
+  inputSchema: MCPTool['inputSchema']
+}
+
+const getEndpointFormat = (endpoint: string): 'openai' | 'anthropic' | 'responses' | null => {
+  if (endpoint.startsWith('/v1/chat/completions')) return 'openai'
+  if (endpoint.startsWith('/v1/messages')) return 'anthropic'
+  if (endpoint.startsWith('/v1/responses')) return 'responses'
+  return null
+}
+
+const buildAssistantModelId = (assistant: AssistantConfig): string | null => {
+  const model = assistant.model ?? assistant.defaultModel
+  if (!model?.provider || !model?.id) {
+    return null
+  }
+  return `${model.provider}:${model.id}`
+}
+
+const applyAssistantMessageOverrides = (
+  body: Record<string, any>,
+  assistant: AssistantConfig,
+  format: 'openai' | 'anthropic' | 'responses'
+): Record<string, any> => {
+  const nextBody = { ...body }
+  const prompt = assistant.prompt ?? ''
+
+  if (format === 'openai') {
+    const messages = Array.isArray(nextBody.messages) ? nextBody.messages : []
+    const filtered = messages.filter((message) => message?.role !== 'system' && message?.role !== 'developer')
+    if (prompt.trim().length > 0) {
+      filtered.unshift({ role: 'system', content: prompt })
+    }
+    nextBody.messages = filtered
+  } else if (format === 'responses') {
+    nextBody.instructions = prompt
+  } else {
+    nextBody.system = prompt
+  }
+
+  return nextBody
+}
+
+const applyAssistantParameterOverrides = (
+  body: Record<string, any>,
+  assistant: AssistantConfig,
+  format: 'openai' | 'anthropic' | 'responses'
+): Record<string, any> => {
+  const nextBody = { ...body }
+  const settings = assistant.settings ?? {}
+
+  if (typeof settings.streamOutput === 'boolean') {
+    nextBody.stream = settings.streamOutput
+  }
+
+  if (settings.enableTemperature && typeof settings.temperature === 'number') {
+    nextBody.temperature = settings.temperature
+  } else if ('temperature' in nextBody) {
+    delete nextBody.temperature
+  }
+
+  if (settings.enableTopP && typeof settings.topP === 'number') {
+    nextBody.top_p = settings.topP
+  } else if ('top_p' in nextBody) {
+    delete nextBody.top_p
+  }
+
+  if (settings.enableMaxTokens && typeof settings.maxTokens === 'number') {
+    if (format === 'responses') {
+      nextBody.max_output_tokens = settings.maxTokens
+      delete nextBody.max_tokens
+    } else {
+      nextBody.max_tokens = settings.maxTokens
+      if ('max_output_tokens' in nextBody) {
+        delete nextBody.max_output_tokens
+      }
+    }
+  } else {
+    if ('max_tokens' in nextBody) {
+      delete nextBody.max_tokens
+    }
+    if ('max_output_tokens' in nextBody) {
+      delete nextBody.max_output_tokens
+    }
+  }
+
+  delete nextBody.tool_choice
+
+  return nextBody
+}
+
+const mapToolsForOpenAI = (tools: ToolDefinition[]) =>
+  tools.map((toolDef) => ({
+    type: 'function',
+    function: {
+      name: toolDef.name,
+      description: toolDef.description || '',
+      parameters: toolDef.inputSchema
+    }
+  }))
+
+const mapToolsForResponses = (tools: ToolDefinition[]) =>
+  tools.map((toolDef) => ({
+    type: 'function',
+    name: toolDef.name,
+    description: toolDef.description || '',
+    parameters: toolDef.inputSchema
+  }))
+
+const mapToolsForAnthropic = (tools: ToolDefinition[]) =>
+  tools.map((toolDef) => ({
+    name: toolDef.name,
+    description: toolDef.description || '',
+    input_schema: toolDef.inputSchema
+  }))
+
+const buildAssistantTools = async (assistant: AssistantConfig): Promise<ToolDefinition[]> => {
+  const serverIds = assistant.mcpServers?.map((server) => server.id).filter(Boolean) ?? []
+  if (serverIds.length === 0) {
+    return []
+  }
+
+  const allowedTools = Array.isArray(assistant.allowed_tools) ? new Set(assistant.allowed_tools) : null
+  const servers = await getMCPServersFromRedux()
+  const tools: ToolDefinition[] = []
+
+  for (const serverId of serverIds) {
+    const server = servers.find((entry) => entry.id === serverId)
+    if (!server || !server.isActive) {
+      continue
+    }
+
+    const info = await mcpApiService.getServerInfo(serverId)
+    if (!info?.tools || !Array.isArray(info.tools)) {
+      continue
+    }
+
+    for (const tool of info.tools as Array<{
+      name: string
+      description?: string
+      inputSchema?: MCPTool['inputSchema']
+    }>) {
+      if (!tool?.name || !tool.inputSchema) {
+        continue
+      }
+
+      if (server.disabledTools?.includes(tool.name)) {
+        continue
+      }
+
+      const toolName = buildFunctionCallToolName(info.name, tool.name)
+      if (allowedTools && !allowedTools.has(toolName)) {
+        continue
+      }
+
+      tools.push({
+        name: toolName,
+        description: tool.description,
+        inputSchema: tool.inputSchema
+      })
+    }
+  }
+
+  return tools
+}
+
+const resolveAssistantById = async (assistantId: string): Promise<AssistantConfig | null> => {
+  const assistants = (await reduxService.select('state.assistants.assistants')) as AssistantConfig[] | null
+  return assistants?.find((assistant) => assistant.id === assistantId) ?? null
+}
+
 /**
  * Gateway middleware for model group routing
  *
@@ -43,16 +240,79 @@ export const gatewayMiddleware = async (req: Request, res: Response, next: NextF
         return
       }
 
-      // Inject the group's model into the request
-      req.body = {
-        ...req.body,
-        model: `${group.providerId}:${group.modelId}`
-      }
+      const endpoint = req.path.startsWith('/') ? req.path : `/${req.path}`
+      const endpointFormat = getEndpointFormat(endpoint)
 
-      logger.debug('Injected model from group', {
-        groupName,
-        model: `${group.providerId}:${group.modelId}`
-      })
+      if (group.mode === 'assistant' && group.assistantId) {
+        if (!endpointFormat) {
+          res.status(400).json({
+            error: {
+              type: 'invalid_request_error',
+              message: `Unsupported endpoint ${endpoint}`
+            }
+          })
+          return
+        }
+
+        const assistant = await resolveAssistantById(group.assistantId)
+        if (!assistant) {
+          res.status(400).json({
+            error: {
+              type: 'invalid_request_error',
+              message: `Assistant '${group.assistantId}' not found`
+            }
+          })
+          return
+        }
+
+        const modelId = buildAssistantModelId(assistant)
+        if (!modelId) {
+          res.status(400).json({
+            error: {
+              type: 'invalid_request_error',
+              message: `Assistant '${group.assistantId}' is missing a model`
+            }
+          })
+          return
+        }
+
+        let nextBody = {
+          ...req.body,
+          model: modelId
+        }
+
+        nextBody = applyAssistantMessageOverrides(nextBody, assistant, endpointFormat)
+        nextBody = applyAssistantParameterOverrides(nextBody, assistant, endpointFormat)
+
+        const tools = await buildAssistantTools(assistant)
+        if (endpointFormat === 'openai') {
+          nextBody.tools = tools.length > 0 ? mapToolsForOpenAI(tools) : undefined
+        } else if (endpointFormat === 'responses') {
+          nextBody.tools = tools.length > 0 ? mapToolsForResponses(tools) : undefined
+        } else {
+          nextBody.tools = tools.length > 0 ? mapToolsForAnthropic(tools) : undefined
+        }
+
+        req.body = nextBody
+
+        logger.debug('Injected assistant preset into request', {
+          groupName,
+          assistantId: assistant.id,
+          model: modelId,
+          toolCount: tools.length
+        })
+      } else {
+        // Inject the group's model into the request
+        req.body = {
+          ...req.body,
+          model: `${group.providerId}:${group.modelId}`
+        }
+
+        logger.debug('Injected model from group', {
+          groupName,
+          model: `${group.providerId}:${group.modelId}`
+        })
+      }
     }
 
     // Get the endpoint path (for group routes, use the part after groupName)
diff --git a/src/renderer/src/i18n/locales/en-us.json b/src/renderer/src/i18n/locales/en-us.json
index feb2a804e9..f73c531cbc 100644
--- a/src/renderer/src/i18n/locales/en-us.json
+++ b/src/renderer/src/i18n/locales/en-us.json
@@ -353,6 +353,13 @@
         "description": "Create model groups with unique URLs for different provider/model combinations",
         "empty": "No model groups configured. Click 'Add Group' to create one.",
         "label": "Model Groups",
+        "mode": {
+          "assistant": "Assistant Preset",
+          "assistantHint": "Assistant preset overrides request parameters.",
+          "assistantPlaceholder": "Select assistant",
+          "label": "Mode",
+          "model": "Direct Model"
+        },
         "namePlaceholder": "Group name"
       },
       "networkAccess": {
diff --git a/src/renderer/src/i18n/locales/zh-cn.json b/src/renderer/src/i18n/locales/zh-cn.json
index da208d2407..f5da50c87e 100644
--- a/src/renderer/src/i18n/locales/zh-cn.json
+++ b/src/renderer/src/i18n/locales/zh-cn.json
@@ -353,6 +353,13 @@
         "description": "创建具有唯一 URL 的模型分组，用于不同的提供商/模型组合",
         "empty": "尚未配置模型分组。点击「添加分组」创建一个。",
         "label": "模型分组",
+        "mode": {
+          "assistant": "助手预设",
+          "assistantHint": "助手预设会覆盖请求参数。",
+          "assistantPlaceholder": "选择助手",
+          "label": "模式",
+          "model": "直接模型"
+        },
         "namePlaceholder": "分组名称"
       },
       "networkAccess": {
diff --git a/src/renderer/src/pages/settings/ToolSettings/ApiServerSettings/ApiServerSettings.tsx b/src/renderer/src/pages/settings/ToolSettings/ApiServerSettings/ApiServerSettings.tsx
index ae8b549086..65a2e3df60 100644
--- a/src/renderer/src/pages/settings/ToolSettings/ApiServerSettings/ApiServerSettings.tsx
+++ b/src/renderer/src/pages/settings/ToolSettings/ApiServerSettings/ApiServerSettings.tsx
@@ -46,6 +46,7 @@ const ApiServerSettings: FC = () => {
 
   // API Gateway state with proper defaults
   const apiServerConfig = useSelector((state: RootState) => state.settings.apiServer)
+  const assistants = useSelector((state: RootState) => state.assistants.assistants)
   const { apiServerRunning, apiServerLoading, startApiServer, stopApiServer, restartApiServer, setApiServerEnabled } =
     useApiServer()
 
@@ -108,6 +109,8 @@ const ApiServerSettings: FC = () => {
       name: `group-${apiServerConfig.modelGroups.length + 1}`, // URL-safe name
       providerId: '',
       modelId: '',
+      mode: 'model',
+      assistantId: '',
       createdAt: Date.now()
     }
     dispatch(addApiGatewayModelGroup(newGroup))
@@ -266,7 +269,13 @@ const ApiServerSettings: FC = () => {
         ) : (
           <ModelGroupList>
             {apiServerConfig.modelGroups.map((group) => (
-              <ModelGroupCard key={group.id} group={group} onUpdate={updateModelGroup} onDelete={deleteModelGroup} />
+              <ModelGroupCard
+                key={group.id}
+                group={group}
+                assistants={assistants}
+                onUpdate={updateModelGroup}
+                onDelete={deleteModelGroup}
+              />
             ))}
           </ModelGroupList>
         )}
@@ -295,6 +304,7 @@ const ApiServerSettings: FC = () => {
 // Model Group Card Component
 interface ModelGroupCardProps {
   group: ModelGroup
+  assistants: RootState['assistants']['assistants']
   onUpdate: (group: ModelGroup) => void
   onDelete: (groupId: string) => void
 }
@@ -305,11 +315,12 @@ const ENV_FORMAT_TO_ENDPOINT: Record<EnvFormat, GatewayEndpoint> = {
   responses: '/v1/responses'
 }
 
-const ModelGroupCard: FC<ModelGroupCardProps> = ({ group, onUpdate, onDelete }) => {
+const ModelGroupCard: FC<ModelGroupCardProps> = ({ group, assistants, onUpdate, onDelete }) => {
   const { t } = useTranslation()
   const { providers } = useProviders()
   const apiServerConfig = useSelector((state: RootState) => state.settings.apiServer)
   const [envFormat, setEnvFormat] = useState<EnvFormat>('openai')
+  const mode = group.mode ?? 'model'
 
   // Reset envFormat when selected endpoint is disabled
   useEffect(() => {
@@ -398,7 +409,36 @@ const ModelGroupCard: FC<ModelGroupCardProps> = ({ group, onUpdate, onDelete })
     })
   }
 
-  const isConfigured = group.providerId && group.modelId
+  const handleModeChange = (nextMode: 'model' | 'assistant') => {
+    if (nextMode === mode) return
+    onUpdate({
+      ...group,
+      mode: nextMode,
+      assistantId: nextMode === 'assistant' ? group.assistantId || '' : '',
+      providerId: nextMode === 'model' ? group.providerId : '',
+      modelId: nextMode === 'model' ? group.modelId : ''
+    })
+  }
+
+  const handleAssistantChange = (assistantId: string | null) => {
+    onUpdate({
+      ...group,
+      assistantId: assistantId || ''
+    })
+  }
+
+  const selectedAssistant = useMemo(() => {
+    if (!group.assistantId) return undefined
+    return assistants.find((assistant) => assistant.id === group.assistantId)
+  }, [assistants, group.assistantId])
+
+  const assistantModelLabel = useMemo(() => {
+    const model = selectedAssistant?.model ?? selectedAssistant?.defaultModel
+    if (!model) return undefined
+    return model.name || model.id
+  }, [selectedAssistant])
+
+  const isConfigured = mode === 'assistant' ? !!group.assistantId : !!(group.providerId && group.modelId)
 
   return (
     <GroupCard $configured={!!isConfigured}>
@@ -427,35 +467,71 @@ const ModelGroupCard: FC<ModelGroupCardProps> = ({ group, onUpdate, onDelete })
       </GroupHeader>
 
       <GroupContent>
+        <ModeRow>
+          <ModeLabel>{t('apiGateway.fields.modelGroups.mode.label', 'Mode')}</ModeLabel>
+          <Segmented
+            size="small"
+            value={mode}
+            onChange={(value) => handleModeChange(value as 'model' | 'assistant')}
+            options={[
+              { label: t('apiGateway.fields.modelGroups.mode.model', 'Direct Model'), value: 'model' },
+              { label: t('apiGateway.fields.modelGroups.mode.assistant', 'Assistant Preset'), value: 'assistant' }
+            ]}
+          />
+        </ModeRow>
         <SelectRow>
-          <StyledSelect
-            value={group.providerId || undefined}
-            onChange={(value) => handleProviderChange((value as string) || null)}
-            placeholder={t('apiGateway.fields.defaultModel.providerPlaceholder')}
-            allowClear
-            showSearch
-            optionFilterProp="label"
-            style={{ flex: 1 }}
-            options={providers.map((p) => ({
-              value: p.id,
-              label: p.isSystem ? getProviderLabel(p.id) : p.name
-            }))}
-          />
-          <StyledSelect
-            value={group.modelId || undefined}
-            onChange={(value) => handleModelChange((value as string) || null)}
-            placeholder={t('apiGateway.fields.defaultModel.modelPlaceholder')}
-            allowClear
-            showSearch
-            optionFilterProp="label"
-            disabled={!group.providerId}
-            style={{ flex: 1 }}
-            options={models.map((m) => ({
-              value: m.id,
-              label: m.name || m.id
-            }))}
-          />
+          {mode === 'assistant' ? (
+            <StyledSelect
+              value={group.assistantId || undefined}
+              onChange={(value) => handleAssistantChange((value as string) || null)}
+              placeholder={t('apiGateway.fields.modelGroups.mode.assistantPlaceholder', 'Select assistant')}
+              allowClear
+              showSearch
+              optionFilterProp="label"
+              style={{ flex: 1 }}
+              options={assistants.map((assistant) => ({
+                value: assistant.id,
+                label: assistant.name
+              }))}
+            />
+          ) : (
+            <>
+              <StyledSelect
+                value={group.providerId || undefined}
+                onChange={(value) => handleProviderChange((value as string) || null)}
+                placeholder={t('apiGateway.fields.defaultModel.providerPlaceholder')}
+                allowClear
+                showSearch
+                optionFilterProp="label"
+                style={{ flex: 1 }}
+                options={providers.map((p) => ({
+                  value: p.id,
+                  label: p.isSystem ? getProviderLabel(p.id) : p.name
+                }))}
+              />
+              <StyledSelect
+                value={group.modelId || undefined}
+                onChange={(value) => handleModelChange((value as string) || null)}
+                placeholder={t('apiGateway.fields.defaultModel.modelPlaceholder')}
+                allowClear
+                showSearch
+                optionFilterProp="label"
+                disabled={!group.providerId}
+                style={{ flex: 1 }}
+                options={models.map((m) => ({
+                  value: m.id,
+                  label: m.name || m.id
+                }))}
+              />
+            </>
+          )}
         </SelectRow>
+        {mode === 'assistant' && (
+          <ModeHint>
+            {t('apiGateway.fields.modelGroups.mode.assistantHint', 'Assistant preset overrides request parameters.')}
+            {assistantModelLabel ? ` (${assistantModelLabel})` : ''}
+          </ModeHint>
+        )}
 
         {isConfigured && (
           <GroupUrlSection>
@@ -819,6 +895,23 @@ const GroupContent = styled.div`
   gap: 12px;
 `
 
+const ModeRow = styled.div`
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 12px;
+`
+
+const ModeLabel = styled.div`
+  font-size: 12px;
+  color: var(--color-text-3);
+`
+
+const ModeHint = styled.div`
+  font-size: 12px;
+  color: var(--color-text-3);
+`
+
 const GroupUrlSection = styled.div`
   display: flex;
   flex-direction: column;
diff --git a/src/renderer/src/types/apiServer.ts b/src/renderer/src/types/apiServer.ts
index 28df8b9cd8..a35a70a407 100644
--- a/src/renderer/src/types/apiServer.ts
+++ b/src/renderer/src/types/apiServer.ts
@@ -12,6 +12,8 @@ export type ModelGroup = {
   name: string // display name
   providerId: string
   modelId: string
+  mode?: 'model' | 'assistant'
+  assistantId?: string
   createdAt: number
 }