@dexto/core 1.4.0 → 1.5.0

package/dist/llm/providers/local/ai-sdk-adapter.cjs

@@ -0,0 +1,308 @@
+"use strict";
+var __create = Object.create;
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __getProtoOf = Object.getPrototypeOf;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __export = (target, all) => {
+  for (var name in all)
+    __defProp(target, name, { get: all[name], enumerable: true });
+};
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(
+  // If the importer is in node compatibility mode or this is not an ESM
+  // file that has been converted to a CommonJS file using a Babel-
+  // compatible transform (i.e. "__esModule" has not been set), then set
+  // "default" to the CommonJS "module.exports" for node compatibility.
+  isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", { value: mod, enumerable: true }) : target,
+  mod
+));
+var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
+var ai_sdk_adapter_exports = {};
+__export(ai_sdk_adapter_exports, {
+  createLocalLanguageModel: () => createLocalLanguageModel
+});
+module.exports = __toCommonJS(ai_sdk_adapter_exports);
+var import_node_llama_provider = require("./node-llama-provider.js");
+var import_errors = require("./errors.js");
+var import_registry = require("./registry.js");
+var fs = __toESM(require("fs"), 1);
+var path = __toESM(require("path"), 1);
+var os = __toESM(require("os"), 1);
+function getModelsDirectory() {
+  return path.join(os.homedir(), ".dexto", "models");
+}
+function getInstalledModelInfo(modelId) {
+  const stateFile = path.join(getModelsDirectory(), "state.json");
+  try {
+    if (!fs.existsSync(stateFile)) {
+      return null;
+    }
+    const content = fs.readFileSync(stateFile, "utf-8");
+    const state = JSON.parse(content);
+    return state.installed[modelId] ?? null;
+  } catch {
+    return null;
+  }
+}
+function getCustomModelFilePath(modelId) {
+  const customModelsFile = path.join(getModelsDirectory(), "custom-models.json");
+  try {
+    if (!fs.existsSync(customModelsFile)) {
+      return null;
+    }
+    const content = fs.readFileSync(customModelsFile, "utf-8");
+    const storage = JSON.parse(content);
+    const customModel = storage.models.find(
+      (m) => m.name === modelId && m.provider === "local" && m.filePath
+    );
+    return customModel?.filePath ?? null;
+  } catch {
+    return null;
+  }
+}
+function createLocalLanguageModel(config) {
+  return new LocalLanguageModel(config);
+}
+class LocalLanguageModel {
+  specificationVersion = "v2";
+  provider = "local";
+  modelId;
+  // Local models don't support URL-based content natively
+  supportedUrls = {};
+  config;
+  session = null;
+  loadedModel = null;
+  initPromise = null;
+  deviceName = "Local";
+  constructor(config) {
+    this.modelId = config.modelId;
+    this.config = config;
+  }
+  /**
+   * Initialize the model lazily on first use.
+   */
+  async ensureInitialized() {
+    if (this.session) {
+      return;
+    }
+    if (this.initPromise) {
+      return this.initPromise;
+    }
+    this.initPromise = this.initialize();
+    return this.initPromise;
+  }
+  async initialize() {
+    const {
+      modelId,
+      modelPath: directPath,
+      contextSize,
+      // Let node-llama-cpp default to "auto" if not specified
+      gpuLayers = -1,
+      threads
+    } = this.config;
+    const isInstalled = await (0, import_node_llama_provider.isNodeLlamaCppInstalled)();
+    if (!isInstalled) {
+      throw import_errors.LocalModelError.nodeLlamaNotInstalled();
+    }
+    let modelPath;
+    if (directPath) {
+      modelPath = directPath;
+    } else {
+      const installedModel = getInstalledModelInfo(modelId);
+      if (installedModel) {
+        modelPath = installedModel.filePath;
+      } else {
+        const customPath = getCustomModelFilePath(modelId);
+        if (customPath) {
+          modelPath = customPath;
+        } else {
+          const registryModel = (0, import_registry.getLocalModelById)(modelId);
+          if (!registryModel) {
+            throw import_errors.LocalModelError.modelNotFound(modelId);
+          }
+          throw import_errors.LocalModelError.modelNotDownloaded(modelId);
+        }
+      }
+    }
+    const loadConfig = {
+      modelPath,
+      gpuLayers
+    };
+    if (contextSize !== void 0) {
+      loadConfig.contextSize = contextSize;
+    }
+    if (threads !== void 0) {
+      loadConfig.threads = threads;
+    }
+    this.loadedModel = await (0, import_node_llama_provider.loadModel)(loadConfig);
+    this.deviceName = this.loadedModel.gpuInfo.deviceName || "Local";
+    this.session = await this.loadedModel.createSession();
+  }
+  /**
+   * Non-streaming text generation (V2 interface).
+   */
+  async doGenerate(options) {
+    await this.ensureInitialized();
+    const prompt = this.formatPrompt(options);
+    const maxTokens = options.maxOutputTokens ?? 1024;
+    const temperature = options.temperature ?? 0.7;
+    const promptOptions = {
+      maxTokens,
+      temperature
+    };
+    if (options.abortSignal) {
+      promptOptions.signal = options.abortSignal;
+    }
+    const response = await this.session.prompt(prompt, promptOptions);
+    const inputTokens = Math.ceil(prompt.length / 4);
+    const outputTokens = Math.ceil(response.length / 4);
+    const content = [{ type: "text", text: response }];
+    const finishReason = "stop";
+    const usage = {
+      inputTokens,
+      outputTokens,
+      totalTokens: inputTokens + outputTokens
+    };
+    const warnings = [];
+    return {
+      content,
+      finishReason,
+      usage,
+      providerMetadata: {
+        local: {
+          device: this.deviceName
+        }
+      },
+      warnings
+    };
+  }
+  /**
+   * Streaming text generation (V2 interface).
+   */
+  async doStream(options) {
+    await this.ensureInitialized();
+    const prompt = this.formatPrompt(options);
+    const maxTokens = options.maxOutputTokens ?? 1024;
+    const temperature = options.temperature ?? 0.7;
+    const inputTokens = Math.ceil(prompt.length / 4);
+    let outputTokens = 0;
+    const session = this.session;
+    const textId = "text-0";
+    const streamPromptOptions = {
+      maxTokens,
+      temperature,
+      onToken: (_token) => {
+      }
+    };
+    if (options.abortSignal) {
+      streamPromptOptions.signal = options.abortSignal;
+    }
+    let controller;
+    const stream = new ReadableStream({
+      async start(ctrl) {
+        controller = ctrl;
+        controller.enqueue({
+          type: "stream-start",
+          warnings: []
+        });
+        controller.enqueue({
+          type: "text-start",
+          id: textId
+        });
+        try {
+          streamPromptOptions.onToken = (token) => {
+            outputTokens += 1;
+            controller.enqueue({
+              type: "text-delta",
+              id: textId,
+              delta: token
+            });
+          };
+          await session.prompt(prompt, streamPromptOptions);
+          controller.enqueue({
+            type: "text-end",
+            id: textId
+          });
+          controller.enqueue({
+            type: "finish",
+            finishReason: "stop",
+            usage: {
+              inputTokens,
+              outputTokens,
+              totalTokens: inputTokens + outputTokens
+            }
+          });
+          controller.close();
+        } catch (error) {
+          if (error instanceof Error && error.name === "AbortError") {
+            controller.enqueue({
+              type: "text-end",
+              id: textId
+            });
+            controller.enqueue({
+              type: "finish",
+              finishReason: "stop",
+              usage: {
+                inputTokens,
+                outputTokens,
+                totalTokens: inputTokens + outputTokens
+              }
+            });
+            controller.close();
+          } else {
+            controller.enqueue({
+              type: "error",
+              error
+            });
+            controller.close();
+          }
+        }
+      }
+    });
+    return {
+      stream
+    };
+  }
+  /**
+   * Format the prompt from AI SDK message format.
+   */
+  formatPrompt(options) {
+    const parts = [];
+    if (options.prompt && Array.isArray(options.prompt)) {
+      for (const message of options.prompt) {
+        if (message.role === "system") {
+          parts.push(`System: ${message.content}`);
+        } else if (message.role === "user") {
+          if (Array.isArray(message.content)) {
+            const textParts = message.content.filter((p) => p.type === "text").map((p) => p.text);
+            if (textParts.length > 0) {
+              parts.push(`User: ${textParts.join("\n")}`);
+            }
+          }
+        } else if (message.role === "assistant") {
+          if (Array.isArray(message.content)) {
+            const textParts = message.content.filter((p) => p.type === "text").map((p) => p.text);
+            if (textParts.length > 0) {
+              parts.push(`Assistant: ${textParts.join("\n")}`);
+            }
+          }
+        }
+      }
+    }
+    parts.push("Assistant:");
+    return parts.join("\n\n");
+  }
+}
+// Annotate the CommonJS export names for ESM import in node:
+0 && (module.exports = {
+  createLocalLanguageModel
+});

package/dist/llm/providers/local/ai-sdk-adapter.d.ts

@@ -0,0 +1,29 @@
+/**
+ * Vercel AI SDK adapter for node-llama-cpp.
+ *
+ * This module creates a LanguageModelV2 implementation that wraps node-llama-cpp,
+ * allowing local GGUF models to be used with the Vercel AI SDK.
+ */
+import type { LanguageModelV2 } from '@ai-sdk/provider';
+/**
+ * Configuration for the local model AI SDK adapter.
+ */
+export interface LocalModelAdapterConfig {
+    /** Model ID from the local registry */
+    modelId: string;
+    /** Direct path to model file (optional, overrides modelId lookup) */
+    modelPath?: string;
+    /** Context window size (default: 4096) */
+    contextSize?: number;
+    /** Number of GPU layers to offload (-1 = all, 0 = CPU only) */
+    gpuLayers?: number;
+    /** Number of CPU threads */
+    threads?: number;
+}
+/**
+ * Create a Vercel AI SDK compatible LanguageModelV2 from a local GGUF model.
+ * This is a synchronous function that returns a LanguageModel with lazy initialization.
+ * The actual model loading happens on first use.
+ */
+export declare function createLocalLanguageModel(config: LocalModelAdapterConfig): LanguageModelV2;
+//# sourceMappingURL=ai-sdk-adapter.d.ts.map

package/dist/llm/providers/local/ai-sdk-adapter.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ai-sdk-adapter.d.ts","sourceRoot":"","sources":["../../../../src/llm/providers/local/ai-sdk-adapter.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAIH,OAAO,KAAK,EACR,eAAe,EAOlB,MAAM,kBAAkB,CAAC;AAa1B;;GAEG;AACH,MAAM,WAAW,uBAAuB;IACpC,uCAAuC;IACvC,OAAO,EAAE,MAAM,CAAC;IAChB,qEAAqE;IACrE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,0CAA0C;IAC1C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,+DAA+D;IAC/D,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,4BAA4B;IAC5B,OAAO,CAAC,EAAE,MAAM,CAAC;CACpB;AA+FD;;;;GAIG;AACH,wBAAgB,wBAAwB,CAAC,MAAM,EAAE,uBAAuB,GAAG,eAAe,CAEzF"}

package/dist/llm/providers/local/ai-sdk-adapter.js

@@ -0,0 +1,278 @@
+import "../../../chunk-PTJYTZNU.js";
+import {
+  loadModel,
+  isNodeLlamaCppInstalled
+} from "./node-llama-provider.js";
+import { LocalModelError } from "./errors.js";
+import { getLocalModelById } from "./registry.js";
+import * as fs from "fs";
+import * as path from "path";
+import * as os from "os";
+function getModelsDirectory() {
+  return path.join(os.homedir(), ".dexto", "models");
+}
+function getInstalledModelInfo(modelId) {
+  const stateFile = path.join(getModelsDirectory(), "state.json");
+  try {
+    if (!fs.existsSync(stateFile)) {
+      return null;
+    }
+    const content = fs.readFileSync(stateFile, "utf-8");
+    const state = JSON.parse(content);
+    return state.installed[modelId] ?? null;
+  } catch {
+    return null;
+  }
+}
+function getCustomModelFilePath(modelId) {
+  const customModelsFile = path.join(getModelsDirectory(), "custom-models.json");
+  try {
+    if (!fs.existsSync(customModelsFile)) {
+      return null;
+    }
+    const content = fs.readFileSync(customModelsFile, "utf-8");
+    const storage = JSON.parse(content);
+    const customModel = storage.models.find(
+      (m) => m.name === modelId && m.provider === "local" && m.filePath
+    );
+    return customModel?.filePath ?? null;
+  } catch {
+    return null;
+  }
+}
+function createLocalLanguageModel(config) {
+  return new LocalLanguageModel(config);
+}
+class LocalLanguageModel {
+  specificationVersion = "v2";
+  provider = "local";
+  modelId;
+  // Local models don't support URL-based content natively
+  supportedUrls = {};
+  config;
+  session = null;
+  loadedModel = null;
+  initPromise = null;
+  deviceName = "Local";
+  constructor(config) {
+    this.modelId = config.modelId;
+    this.config = config;
+  }
+  /**
+   * Initialize the model lazily on first use.
+   */
+  async ensureInitialized() {
+    if (this.session) {
+      return;
+    }
+    if (this.initPromise) {
+      return this.initPromise;
+    }
+    this.initPromise = this.initialize();
+    return this.initPromise;
+  }
+  async initialize() {
+    const {
+      modelId,
+      modelPath: directPath,
+      contextSize,
+      // Let node-llama-cpp default to "auto" if not specified
+      gpuLayers = -1,
+      threads
+    } = this.config;
+    const isInstalled = await isNodeLlamaCppInstalled();
+    if (!isInstalled) {
+      throw LocalModelError.nodeLlamaNotInstalled();
+    }
+    let modelPath;
+    if (directPath) {
+      modelPath = directPath;
+    } else {
+      const installedModel = getInstalledModelInfo(modelId);
+      if (installedModel) {
+        modelPath = installedModel.filePath;
+      } else {
+        const customPath = getCustomModelFilePath(modelId);
+        if (customPath) {
+          modelPath = customPath;
+        } else {
+          const registryModel = getLocalModelById(modelId);
+          if (!registryModel) {
+            throw LocalModelError.modelNotFound(modelId);
+          }
+          throw LocalModelError.modelNotDownloaded(modelId);
+        }
+      }
+    }
+    const loadConfig = {
+      modelPath,
+      gpuLayers
+    };
+    if (contextSize !== void 0) {
+      loadConfig.contextSize = contextSize;
+    }
+    if (threads !== void 0) {
+      loadConfig.threads = threads;
+    }
+    this.loadedModel = await loadModel(loadConfig);
+    this.deviceName = this.loadedModel.gpuInfo.deviceName || "Local";
+    this.session = await this.loadedModel.createSession();
+  }
+  /**
+   * Non-streaming text generation (V2 interface).
+   */
+  async doGenerate(options) {
+    await this.ensureInitialized();
+    const prompt = this.formatPrompt(options);
+    const maxTokens = options.maxOutputTokens ?? 1024;
+    const temperature = options.temperature ?? 0.7;
+    const promptOptions = {
+      maxTokens,
+      temperature
+    };
+    if (options.abortSignal) {
+      promptOptions.signal = options.abortSignal;
+    }
+    const response = await this.session.prompt(prompt, promptOptions);
+    const inputTokens = Math.ceil(prompt.length / 4);
+    const outputTokens = Math.ceil(response.length / 4);
+    const content = [{ type: "text", text: response }];
+    const finishReason = "stop";
+    const usage = {
+      inputTokens,
+      outputTokens,
+      totalTokens: inputTokens + outputTokens
+    };
+    const warnings = [];
+    return {
+      content,
+      finishReason,
+      usage,
+      providerMetadata: {
+        local: {
+          device: this.deviceName
+        }
+      },
+      warnings
+    };
+  }
+  /**
+   * Streaming text generation (V2 interface).
+   */
+  async doStream(options) {
+    await this.ensureInitialized();
+    const prompt = this.formatPrompt(options);
+    const maxTokens = options.maxOutputTokens ?? 1024;
+    const temperature = options.temperature ?? 0.7;
+    const inputTokens = Math.ceil(prompt.length / 4);
+    let outputTokens = 0;
+    const session = this.session;
+    const textId = "text-0";
+    const streamPromptOptions = {
+      maxTokens,
+      temperature,
+      onToken: (_token) => {
+      }
+    };
+    if (options.abortSignal) {
+      streamPromptOptions.signal = options.abortSignal;
+    }
+    let controller;
+    const stream = new ReadableStream({
+      async start(ctrl) {
+        controller = ctrl;
+        controller.enqueue({
+          type: "stream-start",
+          warnings: []
+        });
+        controller.enqueue({
+          type: "text-start",
+          id: textId
+        });
+        try {
+          streamPromptOptions.onToken = (token) => {
+            outputTokens += 1;
+            controller.enqueue({
+              type: "text-delta",
+              id: textId,
+              delta: token
+            });
+          };
+          await session.prompt(prompt, streamPromptOptions);
+          controller.enqueue({
+            type: "text-end",
+            id: textId
+          });
+          controller.enqueue({
+            type: "finish",
+            finishReason: "stop",
+            usage: {
+              inputTokens,
+              outputTokens,
+              totalTokens: inputTokens + outputTokens
+            }
+          });
+          controller.close();
+        } catch (error) {
+          if (error instanceof Error && error.name === "AbortError") {
+            controller.enqueue({
+              type: "text-end",
+              id: textId
+            });
+            controller.enqueue({
+              type: "finish",
+              finishReason: "stop",
+              usage: {
+                inputTokens,
+                outputTokens,
+                totalTokens: inputTokens + outputTokens
+              }
+            });
+            controller.close();
+          } else {
+            controller.enqueue({
+              type: "error",
+              error
+            });
+            controller.close();
+          }
+        }
+      }
+    });
+    return {
+      stream
+    };
+  }
+  /**
+   * Format the prompt from AI SDK message format.
+   */
+  formatPrompt(options) {
+    const parts = [];
+    if (options.prompt && Array.isArray(options.prompt)) {
+      for (const message of options.prompt) {
+        if (message.role === "system") {
+          parts.push(`System: ${message.content}`);
+        } else if (message.role === "user") {
+          if (Array.isArray(message.content)) {
+            const textParts = message.content.filter((p) => p.type === "text").map((p) => p.text);
+            if (textParts.length > 0) {
+              parts.push(`User: ${textParts.join("\n")}`);
+            }
+          }
+        } else if (message.role === "assistant") {
+          if (Array.isArray(message.content)) {
+            const textParts = message.content.filter((p) => p.type === "text").map((p) => p.text);
+            if (textParts.length > 0) {
+              parts.push(`Assistant: ${textParts.join("\n")}`);
+            }
+          }
+        }
+      }
+    }
+    parts.push("Assistant:");
+    return parts.join("\n\n");
+  }
+}
+export {
+  createLocalLanguageModel
+};