模型配置

Author: linyuan.yang

packages/admin/src/i18n/en.ts

@@ -218,6 +218,8 @@ export default {
     thinking: 'Thinking',
     thinking_none: 'Default (off)',
     thinking_budget: 'Budget Tokens',
+    prompt_caching: 'Prompt Caching',
+    prompt_caching_desc: 'Enable prompt caching (saves ~90% input cost for multi-turn conversations)',
   },
   embeddings: {
     add: '+ Add Embedding',
@@ -638,11 +640,11 @@ export default {
   usage: {
     title: 'Token Usage Stats',
     date: 'Date',
-    input_tokens: 'Input',
-    output_tokens: 'Output',
-    total_tokens: 'Total',
-    cache_read: 'Cache Hit',
-    cache_creation: 'Cache Create',
+    input_tokens: 'Input Tokens',
+    output_tokens: 'Output Tokens',
+    total_tokens: 'Total Tokens',
+    cache_read: 'Cache Read Tokens',
+    cache_creation: 'Cache Write Tokens',
     no_data: 'No data',
     last: 'Last',
     total: 'Cumulative',

packages/admin/src/i18n/zh.ts

@@ -218,6 +218,8 @@ export default {
     thinking: '思考模式',
     thinking_none: '默认（关闭）',
     thinking_budget: '思考 Token 预算',
+    prompt_caching: '提示缓存',
+    prompt_caching_desc: '启用 Prompt Caching（多轮对话可节省约 90% 输入费用）',
   },
   embeddings: {
     add: '+ 添加向量模型',
@@ -638,11 +640,11 @@ export default {
   usage: {
     title: 'Token 用量统计',
     date: '日期',
-    input_tokens: '输入',
-    output_tokens: '输出',
-    total_tokens: '合计',
-    cache_read: '缓存命中',
-    cache_creation: '缓存创建',
+    input_tokens: '输入 Token',
+    output_tokens: '输出 Token',
+    total_tokens: '总 Token',
+    cache_read: '缓存读取 Token',
+    cache_creation: '缓存写入 Token',
     no_data: '暂无数据',
     last: '最近',
     total: '累计',

packages/admin/src/views/ModelsView.vue

@@ -19,23 +19,34 @@ const showModal   = ref(false)
 const editingName = ref<string | null>(null)
 const showApiKey  = ref(false)
 const form = ref<ModelConfig>({
-  name: '', provider: ModelProvider.OpenAI, baseURL: '', apiKey: '', model: '', apiVersion: undefined, temperature: undefined, maxTokens: undefined, contextWindow: undefined, thinking: undefined,
+  name: '', provider: ModelProvider.OpenAI, baseURL: '', apiKey: '', model: '', temperature: undefined, maxTokens: undefined, contextWindow: undefined,
 })
 
 const isOllama     = computed(() => form.value.provider === ModelProvider.Ollama)
 const isAnthropic  = computed(() => form.value.provider === ModelProvider.Anthropic)
 const isGemini     = computed(() => form.value.provider === ModelProvider.Gemini || form.value.provider === ModelProvider.GeminiImage)
 
 const thinkingType = computed({
-  get: () => form.value.thinking?.type ?? '',
+  get: () => form.value.anthropic?.thinking?.type ?? '',
   set: (v: string) => {
-    if (!v) { form.value.thinking = undefined; return }
-    form.value.thinking = { type: v as any, ...(v === 'enabled' ? { budgetTokens: form.value.thinking?.budgetTokens ?? 8192 } : {}) }
+    if (!v) { form.value.anthropic = { ...form.value.anthropic, thinking: undefined }; return }
+    form.value.anthropic = {
+      ...form.value.anthropic,
+      thinking: { type: v as any, ...(v === 'enabled' ? { budgetTokens: form.value.anthropic?.thinking?.budgetTokens ?? 8192 } : {}) },
+    }
   },
 })
 const thinkingBudget = computed({
-  get: () => form.value.thinking?.budgetTokens,
-  set: (v: number | undefined) => { if (form.value.thinking) form.value.thinking.budgetTokens = v },
+  get: () => form.value.anthropic?.thinking?.budgetTokens,
+  set: (v: number | undefined) => { if (form.value.anthropic?.thinking) form.value.anthropic.thinking.budgetTokens = v },
+})
+const promptCaching = computed({
+  get: () => form.value.anthropic?.promptCaching ?? false,
+  set: (v: boolean) => { form.value.anthropic = { ...form.value.anthropic, promptCaching: v || undefined } },
+})
+const geminiApiVersion = computed({
+  get: () => form.value.gemini?.apiVersion ?? '',
+  set: (v: string) => { form.value.gemini = { ...form.value.gemini, apiVersion: v || undefined } },
 })
 
 
@@ -76,7 +87,7 @@ function pickModel(m: string) {
 function openAdd() {
   editingName.value = null
   showApiKey.value  = false
-  form.value = { name: '', provider: ModelProvider.OpenAI, baseURL: '', apiKey: '', model: '', apiVersion: undefined, temperature: undefined, maxTokens: undefined, contextWindow: undefined }
+  form.value = { name: '', provider: ModelProvider.OpenAI, baseURL: '', apiKey: '', model: '', temperature: undefined, maxTokens: undefined, contextWindow: undefined }
   showModal.value = true
 }
 
@@ -90,10 +101,11 @@ function openEdit(id: string) {
     baseURL: m.baseURL,
     apiKey: m.apiKey,
     model: m.model,
-    apiVersion: m.apiVersion,
     temperature: m.temperature,
     maxTokens: m.maxTokens,
     contextWindow: m.contextWindow,
+    anthropic: m.anthropic ? { ...m.anthropic } : undefined,
+    gemini: m.gemini ? { ...m.gemini } : undefined,
   }
   showModal.value = true
 }
@@ -108,9 +120,16 @@ async function save() {
     if (body.temperature === undefined || body.temperature === null) delete body.temperature
     if (body.maxTokens === undefined || body.maxTokens === null) delete body.maxTokens
     if (body.contextWindow === undefined || body.contextWindow === null) delete body.contextWindow
-    if (!body.apiVersion) delete body.apiVersion
-    if (!body.thinking) delete body.thinking
-    else if (body.thinking.type !== 'enabled') delete body.thinking.budgetTokens
+    if (body.anthropic) {
+      if (!body.anthropic.thinking) delete body.anthropic.thinking
+      else if (body.anthropic.thinking.type !== 'enabled') delete body.anthropic.thinking.budgetTokens
+      if (!body.anthropic.promptCaching) delete body.anthropic.promptCaching
+      if (!Object.keys(body.anthropic).length) delete body.anthropic
+    }
+    if (body.gemini) {
+      if (!body.gemini.apiVersion) delete body.gemini.apiVersion
+      if (!Object.keys(body.gemini).length) delete body.gemini
+    }
     const id = editingName.value
     const res = id
       ? await apiFetch(`/api/settings/models/${encodeURIComponent(id)}`, 'PUT', body)
@@ -257,7 +276,7 @@ async function refresh() {
           </div>
           <div v-if="isGemini" class="form-group">
             <label>{{ t('models.api_version') }}</label>
-            <input v-model="form.apiVersion" placeholder="v1beta" />
+            <input v-model="geminiApiVersion" placeholder="v1beta" />
           </div>
           <div class="form-group">
             <label>{{ t('models.temperature') }}</label>
@@ -284,6 +303,10 @@ async function refresh() {
             <label>{{ t('models.thinking_budget') }}</label>
             <input v-model.number="thinkingBudget" type="number" step="1024" placeholder="8192" />
           </div>
+          <div v-if="isAnthropic" class="form-group">
+            <label>{{ t('models.prompt_caching') }}</label>
+            <label class="checkbox-label"><input type="checkbox" v-model="promptCaching" /> {{ t('models.prompt_caching_desc') }}</label>
+          </div>
         </div>
         <div class="modal-footer">
           <button class="btn-outline" @click="showModal = false">{{ t('common.cancel') }}</button>

packages/channel.xiaoai/src/mi/account.ts

@@ -6,7 +6,7 @@ const SID = 'micoapi';
 const LOGIN_URL = 'https://account.xiaomi.com/pass/serviceLogin';
 const AUTH_URL = 'https://account.xiaomi.com/pass/serviceLoginAuth2';
 const USER_AGENT =
-  'MiHome/6.0.103 (com.xiaomi.mihome; build:6.0.103.1; Android 14) Alamofire/2.0.1 Channel/stable';
+  'Dalvik/2.1.0 (Linux; U; Android 10; RMX2111 Build/QP1A.190711.020) APP/xiaomi.mico APPV/2004040 MK/Uk1YMjExMQ== PassportSDK/3.8.3 passport-ui/3.8.3';
 
 function md5(str: string): string {
   return crypto.createHash('md5').update(str).digest('hex').toUpperCase();
@@ -26,53 +26,73 @@ function randomDeviceId(): string {
 }
 
 function parseAuthResponse(text: string): any {
-  const cleaned = text.replace('&&&START&&&', '');
+  const cleaned = text
+    .replace('&&&START&&&', '')
+    .replace(/:(\d{9,})/g, ':"$1"');
   return JSON.parse(cleaned);
 }
 
+function buildCookies(account: MiAccount, deviceId: string): string {
+  const parts: string[] = [
+    `userId=${account.userId}`,
+    `deviceId=${deviceId}`,
+    'sdkVersion=3.9',
+  ];
+  if (account.passToken) {
+    parts.push(`passToken=${account.passToken}`);
+  }
+  return parts.join('; ');
+}
+
 export async function login(account: MiAccount): Promise<AuthedAccount> {
   const deviceId = account.deviceId || randomDeviceId();
-  const cookies = `userId=${account.userId}; deviceId=${deviceId}`;
+  const cookies = buildCookies(account, deviceId);
 
   const step1 = await axios.get(LOGIN_URL, {
     params: { sid: SID, _json: 'true', _locale: 'zh_CN' },
     headers: { 'User-Agent': USER_AGENT, Cookie: cookies },
     transformResponse: [(data) => data],
   });
-  const challenge = parseAuthResponse(step1.data);
-
-  const formData = new URLSearchParams({
-    _json: 'true',
-    qs: challenge.qs,
-    sid: SID,
-    _sign: challenge._sign,
-    callback: challenge.callback,
-    user: account.userId,
-    hash: md5(account.password),
-  });
+  let pass = parseAuthResponse(step1.data);
 
-  const step2 = await axios.post(AUTH_URL, formData.toString(), {
-    headers: {
-      'User-Agent': USER_AGENT,
-      'Content-Type': 'application/x-www-form-urlencoded',
-      Cookie: cookies,
-    },
-    transformResponse: [(data) => data],
-  });
-  const authResult = parseAuthResponse(step2.data);
+  if (pass.code !== 0) {
+    if (!account.password) {
+      throw new Error('XiaoAi login failed: password required for re-authentication');
+    }
+    const formData = new URLSearchParams({
+      _json: 'true',
+      qs: pass.qs,
+      sid: SID,
+      _sign: pass._sign,
+      callback: pass.callback,
+      user: account.userId,
+      hash: md5(account.password),
+    });
+
+    const step2 = await axios.post(AUTH_URL, formData.toString(), {
+      headers: {
+        'User-Agent': USER_AGENT,
+        'Content-Type': 'application/x-www-form-urlencoded',
+        Cookie: cookies,
+      },
+      transformResponse: [(data) => data],
+    });
+    pass = parseAuthResponse(step2.data);
+  }
+
+  if (pass.location?.includes('identity/authStart')) {
+    throw new Error('XiaoAi login failed: verification code required, check passToken');
+  }
 
-  if (!authResult.location) {
-    throw new Error(`XiaoAi login failed: ${authResult.desc || 'unknown error'}`);
+  if (!pass.location || !pass.nonce || !pass.ssecurity) {
+    throw new Error(`XiaoAi login failed: ${pass.description || pass.desc || 'invalid credentials'}`);
   }
 
-  const nonce = authResult.nonce;
-  const ssecurity = authResult.ssecurity;
-  const clientSign = sha1(`nonce=${nonce}&${ssecurity}`);
-  const tokenUrl = `${authResult.location}&clientSign=${encodeURIComponent(clientSign)}`;
+  const clientSign = sha1(`nonce=${pass.nonce}&${pass.ssecurity}`);
+  const tokenUrl = `${pass.location}&clientSign=${encodeURIComponent(clientSign)}`;
 
   const step3 = await axios.get(tokenUrl, {
     headers: { 'User-Agent': USER_AGENT },
-    maxRedirects: 0,
     validateStatus: (s) => s >= 200 && s < 400,
   });
 

packages/channel.xiaoai/src/mi/types.ts

@@ -1,6 +1,7 @@
 export interface MiAccount {
   userId: string;
   password: string;
+  passToken?: string;
   serviceToken?: string;
   deviceId?: string;
 }

packages/channel.xiaoai/tsconfig.tsbuildinfo

@@ -31,17 +31,26 @@ export interface ThinkingConfig {
   budgetTokens?: number
 }
 
+export interface AnthropicConfig {
+  thinking?: ThinkingConfig
+  promptCaching?: boolean
+}
+
+export interface GeminiConfig {
+  apiVersion?: string
+}
+
 export interface ModelConfig {
   name: string
   provider: ModelProvider
   baseURL: string
   apiKey: string
   model: string
-  apiVersion?: string
   temperature?: number
   maxTokens?: number
   contextWindow?: number
-  thinking?: ThinkingConfig
+  anthropic?: AnthropicConfig
+  gemini?: GeminiConfig
 }
 
 export enum EmbeddingProvider {

packages/desktop/src-tauri/icons/icon.icns

@@ -12,8 +12,13 @@ import { toChatMessage, toBaseMessages } from "../Saver/messageConverter";
 export class AnthropicModelService implements IModelService {
   private model?: ChatAnthropic;
   private boundModel?: any;
+  private cacheControl?: { type: "ephemeral" };
 
-  constructor(private config: ModelConfig) {}
+  constructor(private config: ModelConfig) {
+    if (config.anthropic?.promptCaching) {
+      this.cacheControl = { type: "ephemeral" };
+    }
+  }
 
   get contextWindow(): number | undefined { return this.config.contextWindow; }
 
@@ -24,7 +29,7 @@ export class AnthropicModelService implements IModelService {
       model: this.config.model,
       temperature: this.config.temperature,
       maxTokens: this.config.maxTokens,
-      ...(this.config.thinking && { thinking: this.config.thinking as any }),
+      ...(this.config.anthropic?.thinking && { thinking: this.config.anthropic.thinking as any }),
     });
   }
 
@@ -36,7 +41,10 @@ export class AnthropicModelService implements IModelService {
   async invoke(prompt: string | ChatMessage[], options?: { signal?: AbortSignal }): Promise<ChatMessage> {
     const m = this.boundModel ?? this.model!;
     const input = typeof prompt === 'string' ? prompt : toBaseMessages(prompt);
-    const result = await m.invoke(input, options?.signal ? { signal: options.signal } : undefined);
+    const result = await m.invoke(input, {
+      ...(options?.signal && { signal: options.signal }),
+      ...(this.cacheControl && { cache_control: this.cacheControl }),
+    });
     return toChatMessage(result);
   }
 
@@ -52,7 +60,10 @@ export class AnthropicModelService implements IModelService {
   async stream(messages: string | ChatMessage[], options?: { signal?: AbortSignal }): Promise<AsyncIterable<ChatMessage>> {
     const m = this.boundModel ?? this.model!;
     const input = typeof messages === 'string' ? messages : toBaseMessages(messages);
-    const lcStream = await m.stream(input, options?.signal ? { signal: options.signal } : undefined);
+    const lcStream = await m.stream(input, {
+      ...(options?.signal && { signal: options.signal }),
+      ...(this.cacheControl && { cache_control: this.cacheControl }),
+    });
     return (async function* () {
       let accumulated: AIMessageChunk | undefined;
       for await (const chunk of lcStream) {

packages/sbot.commons/src/settings.ts

@@ -21,7 +21,7 @@ export class GeminiModelService implements IModelService {
       apiKey: this.config.apiKey,
       baseUrl: this.config.baseURL,
       model: this.config.model,
-      apiVersion: this.config.apiVersion ?? "v1",
+      apiVersion: this.config.gemini?.apiVersion ?? "v1",
     };
     if (this.config.temperature != null) opts.temperature = this.config.temperature;
     if (this.config.maxTokens != null) opts.maxOutputTokens = this.config.maxTokens;