feat(semantic-index): move model to ai dir and preload on config save

Move local embedding model storage from getCacheDir() to getAiDataDir()
(~/.chatlab/ai/models/semantic-index) so the "clear cache" button in
settings no longer wipes downloaded models.

Also trigger model download immediately in the background when a local
model is configured, instead of waiting for the first index build.
Download status (idle/downloading/ready/error) is tracked in
SemanticIndexService, exposed via the SemanticIndexRuntime interface and
worker RPC, and surfaced in the settings UI as a pulsing indicator.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

packages/node-runtime/src/semantic-index/embedder-factory.ts

@@ -16,7 +16,7 @@ import type { EmbeddingProvider } from './embedding/types'
 import type { SemanticIndexConfig } from './config'
 
 export interface EmbedderFactoryDeps {
-  /** 本地模型缓存目录，例如 ~/.chatlab/cache/models/semantic-index */
+  /** 本地模型目录，例如 ~/.chatlab/ai/models/semantic-index */
   modelsCacheDir?: string
   /** auth-profiles 解析，默认走 @openchatlab/config */
   resolveApiKey?: (provider: string, authProfile?: string) => string

packages/node-runtime/src/semantic-index/embedding/api.ts

@@ -89,4 +89,8 @@ export class OpenAICompatibleEmbeddingProvider implements EmbeddingProvider {
     const [vector] = await this.embedDocuments([text])
     return vector
   }
+
+  async preload(): Promise<void> {
+    // API mode: no local model to download
+  }
 }

packages/node-runtime/src/semantic-index/embedding/local.ts

@@ -101,4 +101,8 @@ export class LocalEmbeddingProvider implements EmbeddingProvider {
     const [vector] = await this.embedDocuments([withInstruction])
     return vector
   }
+
+  async preload(): Promise<void> {
+    await this.getExtractor()
+  }
 }

packages/node-runtime/src/semantic-index/embedding/types.ts

@@ -21,4 +21,6 @@ export interface EmbeddingProvider {
   embedDocuments(texts: string[]): Promise<Float32Array[]>
   /** query 向量化（本地模型会加 queryInstruction） */
   embedQuery(text: string): Promise<Float32Array>
+  /** 预热：触发模型下载/加载而不做实际 embedding。API provider 为 no-op。 */
+  preload?(): Promise<void>
 }

packages/node-runtime/src/semantic-index/runtime.ts

@@ -17,6 +17,7 @@ export type MaybePromise<T> = T | Promise<T>
 export interface SemanticIndexRuntime {
   getConfig(): MaybePromise<SemanticIndexConfig>
   setConfig(config: SemanticIndexConfigInput, options?: { apiKey?: string }): MaybePromise<SemanticIndexConfig>
+  getModelStatus(): MaybePromise<'idle' | 'downloading' | 'ready' | 'error'>
   isConfigured(): MaybePromise<boolean>
   hasApiKey(): MaybePromise<boolean>
 

packages/node-runtime/src/semantic-index/service.ts

@@ -29,6 +29,7 @@ import {
 import type { ChunkSource } from './chunker'
 import {
   isKeylessSemanticIndexApiBaseUrl,
+  isSemanticIndexConfigured,
   SemanticIndexConfigStore,
   type SemanticIndexConfig,
   type SemanticIndexConfigInput,
@@ -197,7 +198,7 @@ export const SEMANTIC_INDEX_CONFIG_FILE = 'semantic-index-config.json'
  * 按 PathProvider 约定路径创建 service：
  * - 向量库：{vectorDir}/embedding_index.db
  * - 配置：{aiDataDir}/semantic-index-config.json
- * - 本地模型缓存：{cacheDir}/models/semantic-index
+ * - 本地模型：{aiDataDir}/models/semantic-index（不在 cache 目录，不随清理缓存丢失）
  */
 export function createSemanticIndexService(params: {
   pathProvider: PathProvider
@@ -210,7 +211,7 @@ export function createSemanticIndexService(params: {
   return new SemanticIndexService({
     vectorDbPath: path.join(pathProvider.getVectorDir(), SEMANTIC_INDEX_DB_FILE),
     configPath: path.join(pathProvider.getAiDataDir(), SEMANTIC_INDEX_CONFIG_FILE),
-    modelsCacheDir: path.join(pathProvider.getCacheDir(), 'models', 'semantic-index'),
+    modelsCacheDir: path.join(pathProvider.getAiDataDir(), 'models', 'semantic-index'),
     sessionAdapter: params.sessionAdapter,
     nativeBinding: params.nativeBinding,
     loadSqliteVec: params.loadSqliteVec,
@@ -228,6 +229,7 @@ export class SemanticIndexService {
 
   private embedder: EmbeddingProvider | null = null
   private embedderModelId: string | null = null
+  private modelPreloadStatus: 'idle' | 'downloading' | 'ready' | 'error' = 'idle'
 
   /** 当前激活的 chunker 身份（version + 参数 hash），随索引记录用于重建判定 */
   private readonly chunkerVersion = CHUNKER_VERSION
@@ -277,6 +279,10 @@ export class SemanticIndexService {
    * 保存配置；模型身份变化会使已启用对话的索引在重建完成前不被检索使用。
    * 传入 apiKey 时写入固定 auth profile，config 内只保存引用，绝不持久化明文。
    */
+  getModelStatus(): 'idle' | 'downloading' | 'ready' | 'error' {
+    return this.modelPreloadStatus
+  }
+
   setConfig(config: SemanticIndexConfigInput, options?: { apiKey?: string }): SemanticIndexConfig {
     const saved = persistSemanticIndexConfig(this.configStore, config, {
       apiKey: options?.apiKey,
@@ -284,10 +290,23 @@ export class SemanticIndexService {
     })
     this.embedder = null
     this.embedderModelId = null
+    this.modelPreloadStatus = 'idle'
     // 关闭全局开关时停止所有在跑/排队的建索引任务（保留已建立的部分数据）
     if (!saved.enabled) {
       for (const state of this.stateStore.listEnabled()) this.queue.pause(state.dbPathHash)
     }
+    // 本地模式已配置时，立即在后台触发模型下载，让用户在建索引前完成等待
+    if (isSemanticIndexConfigured(saved) && saved.mode === 'local') {
+      this.modelPreloadStatus = 'downloading'
+      void this.getEmbedder()
+        .preload?.()
+        .then(() => {
+          this.modelPreloadStatus = 'ready'
+        })
+        .catch(() => {
+          this.modelPreloadStatus = 'error'
+        })
+    }
     return saved
   }
 

packages/node-runtime/src/semantic-index/worker-client.ts

@@ -101,6 +101,11 @@ export class SemanticIndexWorkerClient implements SemanticIndexRuntime {
     return resolveSemanticIndexApiKeySet(this.configStore.get(), this.resolveApiKey)
   }
 
+  async getModelStatus(): Promise<'idle' | 'downloading' | 'ready' | 'error'> {
+    if (!this.transport) return 'idle'
+    return this.call<'idle' | 'downloading' | 'ready' | 'error'>('getModelStatus', [])
+  }
+
   enable(sessionId: string): Promise<void> {
     return this.call('enable', [sessionId])
   }