wllama-service 1.0.1 → 2.0.0

package/dist/index.js

@@ -74,7 +74,8 @@ var WllamaService = class {
       return { success: true, usedWebGPU: hasWebGPU };
     } catch (e) {
       await this.unload();
-      return { success: false, error: e.message };
+      const message = e instanceof Error ? e.message : "Failed to load model";
+      return { success: false, error: message };
     }
   }
   /** Generate a response from the loaded model */
@@ -101,7 +102,8 @@ var WllamaService = class {
         timeMs: Math.round(performance.now() - t0)
       };
     } catch (e) {
-      return { success: false, error: e.message || "Generation failed" };
+      const message = e instanceof Error ? e.message : "Generation failed";
+      return { success: false, error: message };
     }
   }
   /** Unload the current model and free memory */
@@ -132,9 +134,12 @@ var WllamaService = class {
   async storeInCache(file) {
     const url = `${window.location.origin}/wllama-local-models/${file.name}`;
     const cache = await caches.open("wllama-local-models");
-    await cache.put(url, new Response(file, {
-      headers: { "Content-Type": "application/octet-stream" }
-    }));
+    await cache.put(
+      url,
+      new Response(file, {
+        headers: { "Content-Type": "application/octet-stream" }
+      })
+    );
   }
   async deleteFromCache(fileName) {
     const url = `${window.location.origin}/wllama-local-models/${fileName}`;

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../src/index.ts","../src/WllamaService.ts"],"sourcesContent":["export { WllamaService } from './WllamaService';\nexport type {\n  WllamaServiceConfig,\n  GenerateRequest,\n  GenerateResult,\n  LoadModelResult,\n  ConnectionResult,\n} from './types';","import { Wllama } from '@wllama/wllama/esm/index.js';\nimport type {\n  WllamaServiceConfig,\n  GenerateRequest,\n  GenerateResult,\n  LoadModelResult,\n  ConnectionResult,\n} from './types';\n\nexport class WllamaService {\n  private wllama: Wllama | null = null;\n  private modelName: string = '';\n  private config: Required<WllamaServiceConfig>;\n\n  constructor(config: WllamaServiceConfig = {}) {\n    this.config = {\n      wasmPath: config.wasmPath ?? '/wllama/wllama.wasm',\n      nGpuLayers: config.nGpuLayers ?? 999,\n      nCtx: config.nCtx ?? 2048,\n    };\n  }\n\n  /** Check browser compatibility and capabilities */\n  checkEnvironment(): ConnectionResult {\n    if (typeof WebAssembly === 'undefined') {\n      return { success: false, error: 'WebAssembly is not supported in this browser.' };\n    }\n    if (!('caches' in window)) {\n      return { success: false, error: 'Cache API is not supported in this browser.' };\n    }\n    const hasWebGPU = !!(navigator as any).gpu;\n    const isolated = window.crossOriginIsolated;\n    return {\n      success: true,\n      hasWebGPU,\n      crossOriginIsolated: isolated,\n    };\n  }\n\n  /** Load a GGUF model file from the user's filesystem */\n  async loadModel(\n    file: File,\n    onProgress?: (progress: number) => void,\n  ): Promise<LoadModelResult> {\n    const env = this.checkEnvironment();\n    if (!env.success) return { success: false, error: env.error };\n\n    await this.unload();\n\n    try {\n      onProgress?.(10);\n\n      const configPaths = { default: this.config.wasmPath };\n      this.wllama = new Wllama(configPaths as any);\n      onProgress?.(30);\n\n      await this.storeInCache(file);\n      onProgress?.(50);\n\n      const hasWebGPU = !!(navigator as any).gpu;\n      await (this.wllama as any).loadModel([file], {\n        n_ctx: this.config.nCtx,\n        ...(hasWebGPU && this.config.nGpuLayers > 0\n          ? { n_gpu_layers: this.config.nGpuLayers }\n          : {}),\n      });\n\n      this.modelName = file.name;\n      onProgress?.(100);\n\n      return { success: true, usedWebGPU: hasWebGPU };\n    } catch (e: any) {\n      await this.unload();\n      return { success: false, error: e.message };\n    }\n  }\n\n  /** Generate a response from the loaded model */\n  async generate(req: GenerateRequest): Promise<GenerateResult> {\n    if (!this.wllama) {\n      return { success: false, error: 'No model loaded. Call loadModel() first.' };\n    }\n\n    try {\n      const t0 = performance.now();\n\n      const messages: { role: string; content: string }[] = [];\n      if (req.system) messages.push({ role: 'system', content: req.system });\n      messages.push({ role: 'user', content: req.prompt });\n\n      const response = await (this.wllama as any).createChatCompletion({\n        messages,\n        max_tokens: req.maxTokens ?? 512,\n        temperature: req.temperature ?? 0.7,\n        top_k: req.topK ?? 40,\n        top_p: req.topP ?? 0.95,\n      });\n\n      const text: string = response?.choices?.[0]?.message?.content ?? '';\n      return {\n        success: true,\n        text,\n        timeMs: Math.round(performance.now() - t0),\n      };\n    } catch (e: any) {\n      return { success: false, error: e.message || 'Generation failed' };\n    }\n  }\n\n  /** Unload the current model and free memory */\n  async unload(): Promise<void> {\n    if (this.wllama) {\n      try { await (this.wllama as any).exit(); } catch (_) {}\n      this.wllama = null;\n    }\n    if (this.modelName) {\n      try { await this.deleteFromCache(this.modelName); } catch (_) {}\n      this.modelName = '';\n    }\n  }\n\n  /** Whether a model is currently loaded */\n  get isLoaded(): boolean {\n    return this.wllama !== null;\n  }\n\n  /** Name of the currently loaded model */\n  get currentModel(): string {\n    return this.modelName;\n  }\n\n  private async storeInCache(file: File): Promise<void> {\n    const url = `${window.location.origin}/wllama-local-models/${file.name}`;\n    const cache = await caches.open('wllama-local-models');\n    await cache.put(url, new Response(file, {\n      headers: { 'Content-Type': 'application/octet-stream' },\n    }));\n  }\n\n  private async deleteFromCache(fileName: string): Promise<void> {\n    const url = `${window.location.origin}/wllama-local-models/${fileName}`;\n    const cache = await caches.open('wllama-local-models');\n    await cache.delete(url);\n  }\n}"],"mappings":";;;;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;;;ACAA,iBAAuB;AAShB,IAAM,gBAAN,MAAoB;AAAA,EAKzB,YAAY,SAA8B,CAAC,GAAG;AAJ9C,SAAQ,SAAwB;AAChC,SAAQ,YAAoB;AAI1B,SAAK,SAAS;AAAA,MACZ,UAAU,OAAO,YAAY;AAAA,MAC7B,YAAY,OAAO,cAAc;AAAA,MACjC,MAAM,OAAO,QAAQ;AAAA,IACvB;AAAA,EACF;AAAA;AAAA,EAGA,mBAAqC;AACnC,QAAI,OAAO,gBAAgB,aAAa;AACtC,aAAO,EAAE,SAAS,OAAO,OAAO,gDAAgD;AAAA,IAClF;AACA,QAAI,EAAE,YAAY,SAAS;AACzB,aAAO,EAAE,SAAS,OAAO,OAAO,8CAA8C;AAAA,IAChF;AACA,UAAM,YAAY,CAAC,CAAE,UAAkB;AACvC,UAAM,WAAW,OAAO;AACxB,WAAO;AAAA,MACL,SAAS;AAAA,MACT;AAAA,MACA,qBAAqB;AAAA,IACvB;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,UACJ,MACA,YAC0B;AAC1B,UAAM,MAAM,KAAK,iBAAiB;AAClC,QAAI,CAAC,IAAI,QAAS,QAAO,EAAE,SAAS,OAAO,OAAO,IAAI,MAAM;AAE5D,UAAM,KAAK,OAAO;AAElB,QAAI;AACF,mBAAa,EAAE;AAEf,YAAM,cAAc,EAAE,SAAS,KAAK,OAAO,SAAS;AACpD,WAAK,SAAS,IAAI,kBAAO,WAAkB;AAC3C,mBAAa,EAAE;AAEf,YAAM,KAAK,aAAa,IAAI;AAC5B,mBAAa,EAAE;AAEf,YAAM,YAAY,CAAC,CAAE,UAAkB;AACvC,YAAO,KAAK,OAAe,UAAU,CAAC,IAAI,GAAG;AAAA,QAC3C,OAAO,KAAK,OAAO;AAAA,QACnB,GAAI,aAAa,KAAK,OAAO,aAAa,IACtC,EAAE,cAAc,KAAK,OAAO,WAAW,IACvC,CAAC;AAAA,MACP,CAAC;AAED,WAAK,YAAY,KAAK;AACtB,mBAAa,GAAG;AAEhB,aAAO,EAAE,SAAS,MAAM,YAAY,UAAU;AAAA,IAChD,SAAS,GAAQ;AACf,YAAM,KAAK,OAAO;AAClB,aAAO,EAAE,SAAS,OAAO,OAAO,EAAE,QAAQ;AAAA,IAC5C;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,SAAS,KAA+C;AAC5D,QAAI,CAAC,KAAK,QAAQ;AAChB,aAAO,EAAE,SAAS,OAAO,OAAO,2CAA2C;AAAA,IAC7E;AAEA,QAAI;AACF,YAAM,KAAK,YAAY,IAAI;AAE3B,YAAM,WAAgD,CAAC;AACvD,UAAI,IAAI,OAAQ,UAAS,KAAK,EAAE,MAAM,UAAU,SAAS,IAAI,OAAO,CAAC;AACrE,eAAS,KAAK,EAAE,MAAM,QAAQ,SAAS,IAAI,OAAO,CAAC;AAEnD,YAAM,WAAW,MAAO,KAAK,OAAe,qBAAqB;AAAA,QAC/D;AAAA,QACA,YAAY,IAAI,aAAa;AAAA,QAC7B,aAAa,IAAI,eAAe;AAAA,QAChC,OAAO,IAAI,QAAQ;AAAA,QACnB,OAAO,IAAI,QAAQ;AAAA,MACrB,CAAC;AAED,YAAM,OAAe,UAAU,UAAU,CAAC,GAAG,SAAS,WAAW;AACjE,aAAO;AAAA,QACL,SAAS;AAAA,QACT;AAAA,QACA,QAAQ,KAAK,MAAM,YAAY,IAAI,IAAI,EAAE;AAAA,MAC3C;AAAA,IACF,SAAS,GAAQ;AACf,aAAO,EAAE,SAAS,OAAO,OAAO,EAAE,WAAW,oBAAoB;AAAA,IACnE;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,SAAwB;AAC5B,QAAI,KAAK,QAAQ;AACf,UAAI;AAAE,cAAO,KAAK,OAAe,KAAK;AAAA,MAAG,SAAS,GAAG;AAAA,MAAC;AACtD,WAAK,SAAS;AAAA,IAChB;AACA,QAAI,KAAK,WAAW;AAClB,UAAI;AAAE,cAAM,KAAK,gBAAgB,KAAK,SAAS;AAAA,MAAG,SAAS,GAAG;AAAA,MAAC;AAC/D,WAAK,YAAY;AAAA,IACnB;AAAA,EACF;AAAA;AAAA,EAGA,IAAI,WAAoB;AACtB,WAAO,KAAK,WAAW;AAAA,EACzB;AAAA;AAAA,EAGA,IAAI,eAAuB;AACzB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAc,aAAa,MAA2B;AACpD,UAAM,MAAM,GAAG,OAAO,SAAS,MAAM,wBAAwB,KAAK,IAAI;AACtE,UAAM,QAAQ,MAAM,OAAO,KAAK,qBAAqB;AACrD,UAAM,MAAM,IAAI,KAAK,IAAI,SAAS,MAAM;AAAA,MACtC,SAAS,EAAE,gBAAgB,2BAA2B;AAAA,IACxD,CAAC,CAAC;AAAA,EACJ;AAAA,EAEA,MAAc,gBAAgB,UAAiC;AAC7D,UAAM,MAAM,GAAG,OAAO,SAAS,MAAM,wBAAwB,QAAQ;AACrE,UAAM,QAAQ,MAAM,OAAO,KAAK,qBAAqB;AACrD,UAAM,MAAM,OAAO,GAAG;AAAA,EACxB;AACF;","names":[]}
+{"version":3,"sources":["../src/index.ts","../src/WllamaService.ts"],"sourcesContent":["export { WllamaService } from './WllamaService';\nexport type {\n  WllamaServiceConfig,\n  GenerateRequest,\n  GenerateResult,\n  LoadModelResult,\n  ConnectionResult,\n} from './types';","import { Wllama } from '@wllama/wllama/esm/index.js';\nimport type {\n  WllamaServiceConfig,\n  GenerateRequest,\n  GenerateResult,\n  LoadModelResult,\n  ConnectionResult,\n} from './types';\n\nexport class WllamaService {\n  private wllama: Wllama | null = null;\n  private modelName: string = '';\n  private config: Required<WllamaServiceConfig>;\n\n  constructor(config: WllamaServiceConfig = {}) {\n    this.config = {\n      wasmPath: config.wasmPath ?? '/wllama/wllama.wasm',\n      nGpuLayers: config.nGpuLayers ?? 999,\n      nCtx: config.nCtx ?? 2048,\n    };\n  }\n\n  /** Check browser compatibility and capabilities */\n  checkEnvironment(): ConnectionResult {\n    if (typeof WebAssembly === 'undefined') {\n      return { success: false, error: 'WebAssembly is not supported in this browser.' };\n    }\n    if (!('caches' in window)) {\n      return { success: false, error: 'Cache API is not supported in this browser.' };\n    }\n    const hasWebGPU = !!(navigator as any).gpu;\n    const isolated = window.crossOriginIsolated;\n    return {\n      success: true,\n      hasWebGPU,\n      crossOriginIsolated: isolated,\n    };\n  }\n\n  /** Load a GGUF model file from the user's filesystem */\n  async loadModel(\n    file: File,\n    onProgress?: (progress: number) => void,\n  ): Promise<LoadModelResult> {\n    const env = this.checkEnvironment();\n    if (!env.success) return { success: false, error: env.error };\n\n    await this.unload();\n\n    try {\n      onProgress?.(10);\n\n      // AssetsPathConfig requires `default` — this is the key the\n      // Wllama runtime actually reads (confirmed against the real\n      // @wllama/wllama@3.5.1 implementation, not just its docs).\n      // No `as any` needed: this shape is type-correct as written.\n      const configPaths = { default: this.config.wasmPath };\n      this.wllama = new Wllama(configPaths);\n      onProgress?.(30);\n\n      await this.storeInCache(file);\n      onProgress?.(50);\n\n      const hasWebGPU = !!(navigator as any).gpu;\n      await (this.wllama as any).loadModel([file], {\n        n_ctx: this.config.nCtx,\n        ...(hasWebGPU && this.config.nGpuLayers > 0\n          ? { n_gpu_layers: this.config.nGpuLayers }\n          : {}),\n      });\n\n      this.modelName = file.name;\n      onProgress?.(100);\n\n      return { success: true, usedWebGPU: hasWebGPU };\n    } catch (e: unknown) {\n      await this.unload();\n      const message = e instanceof Error ? e.message : 'Failed to load model';\n      return { success: false, error: message };\n    }\n  }\n\n  /** Generate a response from the loaded model */\n  async generate(req: GenerateRequest): Promise<GenerateResult> {\n    if (!this.wllama) {\n      return { success: false, error: 'No model loaded. Call loadModel() first.' };\n    }\n\n    try {\n      const t0 = performance.now();\n\n      const messages: { role: string; content: string }[] = [];\n      if (req.system) messages.push({ role: 'system', content: req.system });\n      messages.push({ role: 'user', content: req.prompt });\n\n      const response = await (this.wllama as any).createChatCompletion({\n        messages,\n        max_tokens: req.maxTokens ?? 512,\n        temperature: req.temperature ?? 0.7,\n        top_k: req.topK ?? 40,\n        top_p: req.topP ?? 0.95,\n      });\n\n      const text: string = response?.choices?.[0]?.message?.content ?? '';\n      return {\n        success: true,\n        text,\n        timeMs: Math.round(performance.now() - t0),\n      };\n    } catch (e: unknown) {\n      const message = e instanceof Error ? e.message : 'Generation failed';\n      return { success: false, error: message };\n    }\n  }\n\n  /** Unload the current model and free memory */\n  async unload(): Promise<void> {\n    if (this.wllama) {\n      try {\n        await (this.wllama as any).exit();\n      } catch (_) {\n        /* ignore */\n      }\n      this.wllama = null;\n    }\n    if (this.modelName) {\n      try {\n        await this.deleteFromCache(this.modelName);\n      } catch (_) {\n        /* ignore */\n      }\n      this.modelName = '';\n    }\n  }\n\n  /** Whether a model is currently loaded */\n  get isLoaded(): boolean {\n    return this.wllama !== null;\n  }\n\n  /** Name of the currently loaded model */\n  get currentModel(): string {\n    return this.modelName;\n  }\n\n  private async storeInCache(file: File): Promise<void> {\n    const url = `${window.location.origin}/wllama-local-models/${file.name}`;\n    const cache = await caches.open('wllama-local-models');\n    await cache.put(\n      url,\n      new Response(file, {\n        headers: { 'Content-Type': 'application/octet-stream' },\n      }),\n    );\n  }\n\n  private async deleteFromCache(fileName: string): Promise<void> {\n    const url = `${window.location.origin}/wllama-local-models/${fileName}`;\n    const cache = await caches.open('wllama-local-models');\n    await cache.delete(url);\n  }\n}"],"mappings":";;;;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;;;ACAA,iBAAuB;AAShB,IAAM,gBAAN,MAAoB;AAAA,EAKzB,YAAY,SAA8B,CAAC,GAAG;AAJ9C,SAAQ,SAAwB;AAChC,SAAQ,YAAoB;AAI1B,SAAK,SAAS;AAAA,MACZ,UAAU,OAAO,YAAY;AAAA,MAC7B,YAAY,OAAO,cAAc;AAAA,MACjC,MAAM,OAAO,QAAQ;AAAA,IACvB;AAAA,EACF;AAAA;AAAA,EAGA,mBAAqC;AACnC,QAAI,OAAO,gBAAgB,aAAa;AACtC,aAAO,EAAE,SAAS,OAAO,OAAO,gDAAgD;AAAA,IAClF;AACA,QAAI,EAAE,YAAY,SAAS;AACzB,aAAO,EAAE,SAAS,OAAO,OAAO,8CAA8C;AAAA,IAChF;AACA,UAAM,YAAY,CAAC,CAAE,UAAkB;AACvC,UAAM,WAAW,OAAO;AACxB,WAAO;AAAA,MACL,SAAS;AAAA,MACT;AAAA,MACA,qBAAqB;AAAA,IACvB;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,UACJ,MACA,YAC0B;AAC1B,UAAM,MAAM,KAAK,iBAAiB;AAClC,QAAI,CAAC,IAAI,QAAS,QAAO,EAAE,SAAS,OAAO,OAAO,IAAI,MAAM;AAE5D,UAAM,KAAK,OAAO;AAElB,QAAI;AACF,mBAAa,EAAE;AAMf,YAAM,cAAc,EAAE,SAAS,KAAK,OAAO,SAAS;AACpD,WAAK,SAAS,IAAI,kBAAO,WAAW;AACpC,mBAAa,EAAE;AAEf,YAAM,KAAK,aAAa,IAAI;AAC5B,mBAAa,EAAE;AAEf,YAAM,YAAY,CAAC,CAAE,UAAkB;AACvC,YAAO,KAAK,OAAe,UAAU,CAAC,IAAI,GAAG;AAAA,QAC3C,OAAO,KAAK,OAAO;AAAA,QACnB,GAAI,aAAa,KAAK,OAAO,aAAa,IACtC,EAAE,cAAc,KAAK,OAAO,WAAW,IACvC,CAAC;AAAA,MACP,CAAC;AAED,WAAK,YAAY,KAAK;AACtB,mBAAa,GAAG;AAEhB,aAAO,EAAE,SAAS,MAAM,YAAY,UAAU;AAAA,IAChD,SAAS,GAAY;AACnB,YAAM,KAAK,OAAO;AAClB,YAAM,UAAU,aAAa,QAAQ,EAAE,UAAU;AACjD,aAAO,EAAE,SAAS,OAAO,OAAO,QAAQ;AAAA,IAC1C;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,SAAS,KAA+C;AAC5D,QAAI,CAAC,KAAK,QAAQ;AAChB,aAAO,EAAE,SAAS,OAAO,OAAO,2CAA2C;AAAA,IAC7E;AAEA,QAAI;AACF,YAAM,KAAK,YAAY,IAAI;AAE3B,YAAM,WAAgD,CAAC;AACvD,UAAI,IAAI,OAAQ,UAAS,KAAK,EAAE,MAAM,UAAU,SAAS,IAAI,OAAO,CAAC;AACrE,eAAS,KAAK,EAAE,MAAM,QAAQ,SAAS,IAAI,OAAO,CAAC;AAEnD,YAAM,WAAW,MAAO,KAAK,OAAe,qBAAqB;AAAA,QAC/D;AAAA,QACA,YAAY,IAAI,aAAa;AAAA,QAC7B,aAAa,IAAI,eAAe;AAAA,QAChC,OAAO,IAAI,QAAQ;AAAA,QACnB,OAAO,IAAI,QAAQ;AAAA,MACrB,CAAC;AAED,YAAM,OAAe,UAAU,UAAU,CAAC,GAAG,SAAS,WAAW;AACjE,aAAO;AAAA,QACL,SAAS;AAAA,QACT;AAAA,QACA,QAAQ,KAAK,MAAM,YAAY,IAAI,IAAI,EAAE;AAAA,MAC3C;AAAA,IACF,SAAS,GAAY;AACnB,YAAM,UAAU,aAAa,QAAQ,EAAE,UAAU;AACjD,aAAO,EAAE,SAAS,OAAO,OAAO,QAAQ;AAAA,IAC1C;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,SAAwB;AAC5B,QAAI,KAAK,QAAQ;AACf,UAAI;AACF,cAAO,KAAK,OAAe,KAAK;AAAA,MAClC,SAAS,GAAG;AAAA,MAEZ;AACA,WAAK,SAAS;AAAA,IAChB;AACA,QAAI,KAAK,WAAW;AAClB,UAAI;AACF,cAAM,KAAK,gBAAgB,KAAK,SAAS;AAAA,MAC3C,SAAS,GAAG;AAAA,MAEZ;AACA,WAAK,YAAY;AAAA,IACnB;AAAA,EACF;AAAA;AAAA,EAGA,IAAI,WAAoB;AACtB,WAAO,KAAK,WAAW;AAAA,EACzB;AAAA;AAAA,EAGA,IAAI,eAAuB;AACzB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAc,aAAa,MAA2B;AACpD,UAAM,MAAM,GAAG,OAAO,SAAS,MAAM,wBAAwB,KAAK,IAAI;AACtE,UAAM,QAAQ,MAAM,OAAO,KAAK,qBAAqB;AACrD,UAAM,MAAM;AAAA,MACV;AAAA,MACA,IAAI,SAAS,MAAM;AAAA,QACjB,SAAS,EAAE,gBAAgB,2BAA2B;AAAA,MACxD,CAAC;AAAA,IACH;AAAA,EACF;AAAA,EAEA,MAAc,gBAAgB,UAAiC;AAC7D,UAAM,MAAM,GAAG,OAAO,SAAS,MAAM,wBAAwB,QAAQ;AACrE,UAAM,QAAQ,MAAM,OAAO,KAAK,qBAAqB;AACrD,UAAM,MAAM,OAAO,GAAG;AAAA,EACxB;AACF;","names":[]}

package/dist/index.mjs

@@ -48,7 +48,8 @@ var WllamaService = class {
       return { success: true, usedWebGPU: hasWebGPU };
     } catch (e) {
       await this.unload();
-      return { success: false, error: e.message };
+      const message = e instanceof Error ? e.message : "Failed to load model";
+      return { success: false, error: message };
     }
   }
   /** Generate a response from the loaded model */
@@ -75,7 +76,8 @@ var WllamaService = class {
         timeMs: Math.round(performance.now() - t0)
       };
     } catch (e) {
-      return { success: false, error: e.message || "Generation failed" };
+      const message = e instanceof Error ? e.message : "Generation failed";
+      return { success: false, error: message };
     }
   }
   /** Unload the current model and free memory */
@@ -106,9 +108,12 @@ var WllamaService = class {
   async storeInCache(file) {
     const url = `${window.location.origin}/wllama-local-models/${file.name}`;
     const cache = await caches.open("wllama-local-models");
-    await cache.put(url, new Response(file, {
-      headers: { "Content-Type": "application/octet-stream" }
-    }));
+    await cache.put(
+      url,
+      new Response(file, {
+        headers: { "Content-Type": "application/octet-stream" }
+      })
+    );
   }
   async deleteFromCache(fileName) {
     const url = `${window.location.origin}/wllama-local-models/${fileName}`;

package/dist/index.mjs.map

@@ -1 +1 @@
-{"version":3,"sources":["../src/WllamaService.ts"],"sourcesContent":["import { Wllama } from '@wllama/wllama/esm/index.js';\nimport type {\n  WllamaServiceConfig,\n  GenerateRequest,\n  GenerateResult,\n  LoadModelResult,\n  ConnectionResult,\n} from './types';\n\nexport class WllamaService {\n  private wllama: Wllama | null = null;\n  private modelName: string = '';\n  private config: Required<WllamaServiceConfig>;\n\n  constructor(config: WllamaServiceConfig = {}) {\n    this.config = {\n      wasmPath: config.wasmPath ?? '/wllama/wllama.wasm',\n      nGpuLayers: config.nGpuLayers ?? 999,\n      nCtx: config.nCtx ?? 2048,\n    };\n  }\n\n  /** Check browser compatibility and capabilities */\n  checkEnvironment(): ConnectionResult {\n    if (typeof WebAssembly === 'undefined') {\n      return { success: false, error: 'WebAssembly is not supported in this browser.' };\n    }\n    if (!('caches' in window)) {\n      return { success: false, error: 'Cache API is not supported in this browser.' };\n    }\n    const hasWebGPU = !!(navigator as any).gpu;\n    const isolated = window.crossOriginIsolated;\n    return {\n      success: true,\n      hasWebGPU,\n      crossOriginIsolated: isolated,\n    };\n  }\n\n  /** Load a GGUF model file from the user's filesystem */\n  async loadModel(\n    file: File,\n    onProgress?: (progress: number) => void,\n  ): Promise<LoadModelResult> {\n    const env = this.checkEnvironment();\n    if (!env.success) return { success: false, error: env.error };\n\n    await this.unload();\n\n    try {\n      onProgress?.(10);\n\n      const configPaths = { default: this.config.wasmPath };\n      this.wllama = new Wllama(configPaths as any);\n      onProgress?.(30);\n\n      await this.storeInCache(file);\n      onProgress?.(50);\n\n      const hasWebGPU = !!(navigator as any).gpu;\n      await (this.wllama as any).loadModel([file], {\n        n_ctx: this.config.nCtx,\n        ...(hasWebGPU && this.config.nGpuLayers > 0\n          ? { n_gpu_layers: this.config.nGpuLayers }\n          : {}),\n      });\n\n      this.modelName = file.name;\n      onProgress?.(100);\n\n      return { success: true, usedWebGPU: hasWebGPU };\n    } catch (e: any) {\n      await this.unload();\n      return { success: false, error: e.message };\n    }\n  }\n\n  /** Generate a response from the loaded model */\n  async generate(req: GenerateRequest): Promise<GenerateResult> {\n    if (!this.wllama) {\n      return { success: false, error: 'No model loaded. Call loadModel() first.' };\n    }\n\n    try {\n      const t0 = performance.now();\n\n      const messages: { role: string; content: string }[] = [];\n      if (req.system) messages.push({ role: 'system', content: req.system });\n      messages.push({ role: 'user', content: req.prompt });\n\n      const response = await (this.wllama as any).createChatCompletion({\n        messages,\n        max_tokens: req.maxTokens ?? 512,\n        temperature: req.temperature ?? 0.7,\n        top_k: req.topK ?? 40,\n        top_p: req.topP ?? 0.95,\n      });\n\n      const text: string = response?.choices?.[0]?.message?.content ?? '';\n      return {\n        success: true,\n        text,\n        timeMs: Math.round(performance.now() - t0),\n      };\n    } catch (e: any) {\n      return { success: false, error: e.message || 'Generation failed' };\n    }\n  }\n\n  /** Unload the current model and free memory */\n  async unload(): Promise<void> {\n    if (this.wllama) {\n      try { await (this.wllama as any).exit(); } catch (_) {}\n      this.wllama = null;\n    }\n    if (this.modelName) {\n      try { await this.deleteFromCache(this.modelName); } catch (_) {}\n      this.modelName = '';\n    }\n  }\n\n  /** Whether a model is currently loaded */\n  get isLoaded(): boolean {\n    return this.wllama !== null;\n  }\n\n  /** Name of the currently loaded model */\n  get currentModel(): string {\n    return this.modelName;\n  }\n\n  private async storeInCache(file: File): Promise<void> {\n    const url = `${window.location.origin}/wllama-local-models/${file.name}`;\n    const cache = await caches.open('wllama-local-models');\n    await cache.put(url, new Response(file, {\n      headers: { 'Content-Type': 'application/octet-stream' },\n    }));\n  }\n\n  private async deleteFromCache(fileName: string): Promise<void> {\n    const url = `${window.location.origin}/wllama-local-models/${fileName}`;\n    const cache = await caches.open('wllama-local-models');\n    await cache.delete(url);\n  }\n}"],"mappings":";AAAA,SAAS,cAAc;AAShB,IAAM,gBAAN,MAAoB;AAAA,EAKzB,YAAY,SAA8B,CAAC,GAAG;AAJ9C,SAAQ,SAAwB;AAChC,SAAQ,YAAoB;AAI1B,SAAK,SAAS;AAAA,MACZ,UAAU,OAAO,YAAY;AAAA,MAC7B,YAAY,OAAO,cAAc;AAAA,MACjC,MAAM,OAAO,QAAQ;AAAA,IACvB;AAAA,EACF;AAAA;AAAA,EAGA,mBAAqC;AACnC,QAAI,OAAO,gBAAgB,aAAa;AACtC,aAAO,EAAE,SAAS,OAAO,OAAO,gDAAgD;AAAA,IAClF;AACA,QAAI,EAAE,YAAY,SAAS;AACzB,aAAO,EAAE,SAAS,OAAO,OAAO,8CAA8C;AAAA,IAChF;AACA,UAAM,YAAY,CAAC,CAAE,UAAkB;AACvC,UAAM,WAAW,OAAO;AACxB,WAAO;AAAA,MACL,SAAS;AAAA,MACT;AAAA,MACA,qBAAqB;AAAA,IACvB;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,UACJ,MACA,YAC0B;AAC1B,UAAM,MAAM,KAAK,iBAAiB;AAClC,QAAI,CAAC,IAAI,QAAS,QAAO,EAAE,SAAS,OAAO,OAAO,IAAI,MAAM;AAE5D,UAAM,KAAK,OAAO;AAElB,QAAI;AACF,mBAAa,EAAE;AAEf,YAAM,cAAc,EAAE,SAAS,KAAK,OAAO,SAAS;AACpD,WAAK,SAAS,IAAI,OAAO,WAAkB;AAC3C,mBAAa,EAAE;AAEf,YAAM,KAAK,aAAa,IAAI;AAC5B,mBAAa,EAAE;AAEf,YAAM,YAAY,CAAC,CAAE,UAAkB;AACvC,YAAO,KAAK,OAAe,UAAU,CAAC,IAAI,GAAG;AAAA,QAC3C,OAAO,KAAK,OAAO;AAAA,QACnB,GAAI,aAAa,KAAK,OAAO,aAAa,IACtC,EAAE,cAAc,KAAK,OAAO,WAAW,IACvC,CAAC;AAAA,MACP,CAAC;AAED,WAAK,YAAY,KAAK;AACtB,mBAAa,GAAG;AAEhB,aAAO,EAAE,SAAS,MAAM,YAAY,UAAU;AAAA,IAChD,SAAS,GAAQ;AACf,YAAM,KAAK,OAAO;AAClB,aAAO,EAAE,SAAS,OAAO,OAAO,EAAE,QAAQ;AAAA,IAC5C;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,SAAS,KAA+C;AAC5D,QAAI,CAAC,KAAK,QAAQ;AAChB,aAAO,EAAE,SAAS,OAAO,OAAO,2CAA2C;AAAA,IAC7E;AAEA,QAAI;AACF,YAAM,KAAK,YAAY,IAAI;AAE3B,YAAM,WAAgD,CAAC;AACvD,UAAI,IAAI,OAAQ,UAAS,KAAK,EAAE,MAAM,UAAU,SAAS,IAAI,OAAO,CAAC;AACrE,eAAS,KAAK,EAAE,MAAM,QAAQ,SAAS,IAAI,OAAO,CAAC;AAEnD,YAAM,WAAW,MAAO,KAAK,OAAe,qBAAqB;AAAA,QAC/D;AAAA,QACA,YAAY,IAAI,aAAa;AAAA,QAC7B,aAAa,IAAI,eAAe;AAAA,QAChC,OAAO,IAAI,QAAQ;AAAA,QACnB,OAAO,IAAI,QAAQ;AAAA,MACrB,CAAC;AAED,YAAM,OAAe,UAAU,UAAU,CAAC,GAAG,SAAS,WAAW;AACjE,aAAO;AAAA,QACL,SAAS;AAAA,QACT;AAAA,QACA,QAAQ,KAAK,MAAM,YAAY,IAAI,IAAI,EAAE;AAAA,MAC3C;AAAA,IACF,SAAS,GAAQ;AACf,aAAO,EAAE,SAAS,OAAO,OAAO,EAAE,WAAW,oBAAoB;AAAA,IACnE;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,SAAwB;AAC5B,QAAI,KAAK,QAAQ;AACf,UAAI;AAAE,cAAO,KAAK,OAAe,KAAK;AAAA,MAAG,SAAS,GAAG;AAAA,MAAC;AACtD,WAAK,SAAS;AAAA,IAChB;AACA,QAAI,KAAK,WAAW;AAClB,UAAI;AAAE,cAAM,KAAK,gBAAgB,KAAK,SAAS;AAAA,MAAG,SAAS,GAAG;AAAA,MAAC;AAC/D,WAAK,YAAY;AAAA,IACnB;AAAA,EACF;AAAA;AAAA,EAGA,IAAI,WAAoB;AACtB,WAAO,KAAK,WAAW;AAAA,EACzB;AAAA;AAAA,EAGA,IAAI,eAAuB;AACzB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAc,aAAa,MAA2B;AACpD,UAAM,MAAM,GAAG,OAAO,SAAS,MAAM,wBAAwB,KAAK,IAAI;AACtE,UAAM,QAAQ,MAAM,OAAO,KAAK,qBAAqB;AACrD,UAAM,MAAM,IAAI,KAAK,IAAI,SAAS,MAAM;AAAA,MACtC,SAAS,EAAE,gBAAgB,2BAA2B;AAAA,IACxD,CAAC,CAAC;AAAA,EACJ;AAAA,EAEA,MAAc,gBAAgB,UAAiC;AAC7D,UAAM,MAAM,GAAG,OAAO,SAAS,MAAM,wBAAwB,QAAQ;AACrE,UAAM,QAAQ,MAAM,OAAO,KAAK,qBAAqB;AACrD,UAAM,MAAM,OAAO,GAAG;AAAA,EACxB;AACF;","names":[]}
+{"version":3,"sources":["../src/WllamaService.ts"],"sourcesContent":["import { Wllama } from '@wllama/wllama/esm/index.js';\nimport type {\n  WllamaServiceConfig,\n  GenerateRequest,\n  GenerateResult,\n  LoadModelResult,\n  ConnectionResult,\n} from './types';\n\nexport class WllamaService {\n  private wllama: Wllama | null = null;\n  private modelName: string = '';\n  private config: Required<WllamaServiceConfig>;\n\n  constructor(config: WllamaServiceConfig = {}) {\n    this.config = {\n      wasmPath: config.wasmPath ?? '/wllama/wllama.wasm',\n      nGpuLayers: config.nGpuLayers ?? 999,\n      nCtx: config.nCtx ?? 2048,\n    };\n  }\n\n  /** Check browser compatibility and capabilities */\n  checkEnvironment(): ConnectionResult {\n    if (typeof WebAssembly === 'undefined') {\n      return { success: false, error: 'WebAssembly is not supported in this browser.' };\n    }\n    if (!('caches' in window)) {\n      return { success: false, error: 'Cache API is not supported in this browser.' };\n    }\n    const hasWebGPU = !!(navigator as any).gpu;\n    const isolated = window.crossOriginIsolated;\n    return {\n      success: true,\n      hasWebGPU,\n      crossOriginIsolated: isolated,\n    };\n  }\n\n  /** Load a GGUF model file from the user's filesystem */\n  async loadModel(\n    file: File,\n    onProgress?: (progress: number) => void,\n  ): Promise<LoadModelResult> {\n    const env = this.checkEnvironment();\n    if (!env.success) return { success: false, error: env.error };\n\n    await this.unload();\n\n    try {\n      onProgress?.(10);\n\n      // AssetsPathConfig requires `default` — this is the key the\n      // Wllama runtime actually reads (confirmed against the real\n      // @wllama/wllama@3.5.1 implementation, not just its docs).\n      // No `as any` needed: this shape is type-correct as written.\n      const configPaths = { default: this.config.wasmPath };\n      this.wllama = new Wllama(configPaths);\n      onProgress?.(30);\n\n      await this.storeInCache(file);\n      onProgress?.(50);\n\n      const hasWebGPU = !!(navigator as any).gpu;\n      await (this.wllama as any).loadModel([file], {\n        n_ctx: this.config.nCtx,\n        ...(hasWebGPU && this.config.nGpuLayers > 0\n          ? { n_gpu_layers: this.config.nGpuLayers }\n          : {}),\n      });\n\n      this.modelName = file.name;\n      onProgress?.(100);\n\n      return { success: true, usedWebGPU: hasWebGPU };\n    } catch (e: unknown) {\n      await this.unload();\n      const message = e instanceof Error ? e.message : 'Failed to load model';\n      return { success: false, error: message };\n    }\n  }\n\n  /** Generate a response from the loaded model */\n  async generate(req: GenerateRequest): Promise<GenerateResult> {\n    if (!this.wllama) {\n      return { success: false, error: 'No model loaded. Call loadModel() first.' };\n    }\n\n    try {\n      const t0 = performance.now();\n\n      const messages: { role: string; content: string }[] = [];\n      if (req.system) messages.push({ role: 'system', content: req.system });\n      messages.push({ role: 'user', content: req.prompt });\n\n      const response = await (this.wllama as any).createChatCompletion({\n        messages,\n        max_tokens: req.maxTokens ?? 512,\n        temperature: req.temperature ?? 0.7,\n        top_k: req.topK ?? 40,\n        top_p: req.topP ?? 0.95,\n      });\n\n      const text: string = response?.choices?.[0]?.message?.content ?? '';\n      return {\n        success: true,\n        text,\n        timeMs: Math.round(performance.now() - t0),\n      };\n    } catch (e: unknown) {\n      const message = e instanceof Error ? e.message : 'Generation failed';\n      return { success: false, error: message };\n    }\n  }\n\n  /** Unload the current model and free memory */\n  async unload(): Promise<void> {\n    if (this.wllama) {\n      try {\n        await (this.wllama as any).exit();\n      } catch (_) {\n        /* ignore */\n      }\n      this.wllama = null;\n    }\n    if (this.modelName) {\n      try {\n        await this.deleteFromCache(this.modelName);\n      } catch (_) {\n        /* ignore */\n      }\n      this.modelName = '';\n    }\n  }\n\n  /** Whether a model is currently loaded */\n  get isLoaded(): boolean {\n    return this.wllama !== null;\n  }\n\n  /** Name of the currently loaded model */\n  get currentModel(): string {\n    return this.modelName;\n  }\n\n  private async storeInCache(file: File): Promise<void> {\n    const url = `${window.location.origin}/wllama-local-models/${file.name}`;\n    const cache = await caches.open('wllama-local-models');\n    await cache.put(\n      url,\n      new Response(file, {\n        headers: { 'Content-Type': 'application/octet-stream' },\n      }),\n    );\n  }\n\n  private async deleteFromCache(fileName: string): Promise<void> {\n    const url = `${window.location.origin}/wllama-local-models/${fileName}`;\n    const cache = await caches.open('wllama-local-models');\n    await cache.delete(url);\n  }\n}"],"mappings":";AAAA,SAAS,cAAc;AAShB,IAAM,gBAAN,MAAoB;AAAA,EAKzB,YAAY,SAA8B,CAAC,GAAG;AAJ9C,SAAQ,SAAwB;AAChC,SAAQ,YAAoB;AAI1B,SAAK,SAAS;AAAA,MACZ,UAAU,OAAO,YAAY;AAAA,MAC7B,YAAY,OAAO,cAAc;AAAA,MACjC,MAAM,OAAO,QAAQ;AAAA,IACvB;AAAA,EACF;AAAA;AAAA,EAGA,mBAAqC;AACnC,QAAI,OAAO,gBAAgB,aAAa;AACtC,aAAO,EAAE,SAAS,OAAO,OAAO,gDAAgD;AAAA,IAClF;AACA,QAAI,EAAE,YAAY,SAAS;AACzB,aAAO,EAAE,SAAS,OAAO,OAAO,8CAA8C;AAAA,IAChF;AACA,UAAM,YAAY,CAAC,CAAE,UAAkB;AACvC,UAAM,WAAW,OAAO;AACxB,WAAO;AAAA,MACL,SAAS;AAAA,MACT;AAAA,MACA,qBAAqB;AAAA,IACvB;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,UACJ,MACA,YAC0B;AAC1B,UAAM,MAAM,KAAK,iBAAiB;AAClC,QAAI,CAAC,IAAI,QAAS,QAAO,EAAE,SAAS,OAAO,OAAO,IAAI,MAAM;AAE5D,UAAM,KAAK,OAAO;AAElB,QAAI;AACF,mBAAa,EAAE;AAMf,YAAM,cAAc,EAAE,SAAS,KAAK,OAAO,SAAS;AACpD,WAAK,SAAS,IAAI,OAAO,WAAW;AACpC,mBAAa,EAAE;AAEf,YAAM,KAAK,aAAa,IAAI;AAC5B,mBAAa,EAAE;AAEf,YAAM,YAAY,CAAC,CAAE,UAAkB;AACvC,YAAO,KAAK,OAAe,UAAU,CAAC,IAAI,GAAG;AAAA,QAC3C,OAAO,KAAK,OAAO;AAAA,QACnB,GAAI,aAAa,KAAK,OAAO,aAAa,IACtC,EAAE,cAAc,KAAK,OAAO,WAAW,IACvC,CAAC;AAAA,MACP,CAAC;AAED,WAAK,YAAY,KAAK;AACtB,mBAAa,GAAG;AAEhB,aAAO,EAAE,SAAS,MAAM,YAAY,UAAU;AAAA,IAChD,SAAS,GAAY;AACnB,YAAM,KAAK,OAAO;AAClB,YAAM,UAAU,aAAa,QAAQ,EAAE,UAAU;AACjD,aAAO,EAAE,SAAS,OAAO,OAAO,QAAQ;AAAA,IAC1C;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,SAAS,KAA+C;AAC5D,QAAI,CAAC,KAAK,QAAQ;AAChB,aAAO,EAAE,SAAS,OAAO,OAAO,2CAA2C;AAAA,IAC7E;AAEA,QAAI;AACF,YAAM,KAAK,YAAY,IAAI;AAE3B,YAAM,WAAgD,CAAC;AACvD,UAAI,IAAI,OAAQ,UAAS,KAAK,EAAE,MAAM,UAAU,SAAS,IAAI,OAAO,CAAC;AACrE,eAAS,KAAK,EAAE,MAAM,QAAQ,SAAS,IAAI,OAAO,CAAC;AAEnD,YAAM,WAAW,MAAO,KAAK,OAAe,qBAAqB;AAAA,QAC/D;AAAA,QACA,YAAY,IAAI,aAAa;AAAA,QAC7B,aAAa,IAAI,eAAe;AAAA,QAChC,OAAO,IAAI,QAAQ;AAAA,QACnB,OAAO,IAAI,QAAQ;AAAA,MACrB,CAAC;AAED,YAAM,OAAe,UAAU,UAAU,CAAC,GAAG,SAAS,WAAW;AACjE,aAAO;AAAA,QACL,SAAS;AAAA,QACT;AAAA,QACA,QAAQ,KAAK,MAAM,YAAY,IAAI,IAAI,EAAE;AAAA,MAC3C;AAAA,IACF,SAAS,GAAY;AACnB,YAAM,UAAU,aAAa,QAAQ,EAAE,UAAU;AACjD,aAAO,EAAE,SAAS,OAAO,OAAO,QAAQ;AAAA,IAC1C;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,SAAwB;AAC5B,QAAI,KAAK,QAAQ;AACf,UAAI;AACF,cAAO,KAAK,OAAe,KAAK;AAAA,MAClC,SAAS,GAAG;AAAA,MAEZ;AACA,WAAK,SAAS;AAAA,IAChB;AACA,QAAI,KAAK,WAAW;AAClB,UAAI;AACF,cAAM,KAAK,gBAAgB,KAAK,SAAS;AAAA,MAC3C,SAAS,GAAG;AAAA,MAEZ;AACA,WAAK,YAAY;AAAA,IACnB;AAAA,EACF;AAAA;AAAA,EAGA,IAAI,WAAoB;AACtB,WAAO,KAAK,WAAW;AAAA,EACzB;AAAA;AAAA,EAGA,IAAI,eAAuB;AACzB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAc,aAAa,MAA2B;AACpD,UAAM,MAAM,GAAG,OAAO,SAAS,MAAM,wBAAwB,KAAK,IAAI;AACtE,UAAM,QAAQ,MAAM,OAAO,KAAK,qBAAqB;AACrD,UAAM,MAAM;AAAA,MACV;AAAA,MACA,IAAI,SAAS,MAAM;AAAA,QACjB,SAAS,EAAE,gBAAgB,2BAA2B;AAAA,MACxD,CAAC;AAAA,IACH;AAAA,EACF;AAAA,EAEA,MAAc,gBAAgB,UAAiC;AAC7D,UAAM,MAAM,GAAG,OAAO,SAAS,MAAM,wBAAwB,QAAQ;AACrE,UAAM,QAAQ,MAAM,OAAO,KAAK,qBAAqB;AACrD,UAAM,MAAM,OAAO,GAAG;AAAA,EACxB;AACF;","names":[]}

package/package.json

@@ -1,8 +1,11 @@
 {
   "name": "wllama-service",
-  "version": "1.0.1",
+  "version": "2.0.0",
   "description": "Framework-agnostic browser LLM service wrapper with WebGPU and WebAssembly support",
   "main": "./dist/index.js",
+  "bin": {
+    "copy-wllama-wasm": "./scripts/postinstall.js"
+  },
   "module": "./dist/index.mjs",
   "types": "./dist/index.d.ts",
   "exports": {
@@ -14,17 +17,13 @@
   },
   "files": [
     "dist",
-    "scripts",
     "README.md"
   ],
   "scripts": {
     "build": "tsup",
     "dev": "tsup --watch",
-    "prepublishOnly": "npm run build",
-    "postinstall": "node scripts/postinstall.js"
-  },
-  "bin": {
-    "wllama-copy-wasm": "./scripts/postinstall.js"
+    "typecheck": "tsc --noEmit",
+    "prepublishOnly": "npm run build"
   },
   "keywords": [
     "wllama",
@@ -36,11 +35,10 @@
     "gguf"
   ],
   "license": "MIT",
-  "peerDependencies": {
-    "@wllama/wllama": ">=3.0.0"
+  "dependencies": {
+    "@wllama/wllama": "^3.5.1"
   },
   "devDependencies": {
-    "@wllama/wllama": "^3.0.0",
     "tsup": "^8.0.0",
     "typescript": "^5.0.0"
   }

package/scripts/postinstall.js

@@ -1,31 +1,26 @@
 #!/usr/bin/env node
-const fs = require('fs');
+'use strict';
+
 const path = require('path');
+const { copyWasm } = require('./copyWllamaWasm');
 
-// Find the consuming app's public directory
-const possiblePublicDirs = [
-  path.resolve(process.cwd(), '../../public'),      // monorepo
-  path.resolve(process.cwd(), '../../../public'),   // deeper monorepo
-  path.resolve(process.cwd(), 'public'),            // root install
-];
+// This file is wired up as the "postinstall" script in the CONSUMING
+// APP's package.json (see setup below) — npm runs it automatically
+// right after `npm install` finishes. No command to type, no step
+// to remember.
+//
+// Destination defaults to <app>/public/wllama, but can be overridden
+// with an env var if the app's public folder lives somewhere else:
+//   WLLAMA_WASM_DEST=./static/wllama npm install
 
-const wasmSrc = path.resolve(
-  __dirname,
-  '../node_modules/@wllama/wllama/esm/wasm/wllama.wasm'
-);
+const destArg = process.env.WLLAMA_WASM_DEST || 'public/wllama';
+const destDir = path.resolve(process.cwd(), destArg);
 
-if (!fs.existsSync(wasmSrc)) {
-  console.warn('[wllama-service] Could not find wllama.wasm — run copy manually');
-  process.exit(0);
-}
+const result = copyWasm(destDir);
 
-const publicDir = possiblePublicDirs.find(fs.existsSync);
-if (!publicDir) {
-  console.warn('[wllama-service] Could not find public/ directory — copy wllama.wasm manually');
+if (!result) {
+  // Non-fatal: don't break `npm install` for the app if something's
+  // off. Worst case, loadModel() fails clearly at runtime with a 404,
+  // which is easier to debug than a broken install.
   process.exit(0);
-}
-
-const dest = path.join(publicDir, 'wllama', 'wllama.wasm');
-fs.mkdirSync(path.dirname(dest), { recursive: true });
-fs.copyFileSync(wasmSrc, dest);
-console.log(`[wllama-service] ✓ Copied wllama.wasm to ${dest}`);
+}