@almadar/integrations 2.22.0 → 2.23.0

package/dist/{contracts-CYtxZXrR.d.ts → contracts-CLTh6gjT.d.ts}

@@ -258,6 +258,15 @@ type LLMIntegrationActions = {
             keyPoints: string[];
         };
     };
+    embed: {
+        params: {
+            texts: string[];
+            model?: string;
+        };
+        result: {
+            embeddings: number[][];
+        };
+    };
 };
 type MLActions = {
     infer: {

package/dist/index.d.ts

@@ -1,6 +1,6 @@
 import { c as IntegrationLogger, b as IntegrationErrorCode, I as IntegrationConfig, B as BaseIntegration, e as IntegrationParams, f as IntegrationResult } from './BaseIntegration-MA-b4fh8.js';
 export { a as IntegrationError, d as IntegrationParamValue, V as ValidationError, g as ValidationResult, v as validateParams } from './BaseIntegration-MA-b4fh8.js';
-export { A as ArxivActions, C as CLIActions, D as DatabaseActions, a as DatabaseDriver, b as DatabaseQueryParamValue, c as DatabaseQueryParams, d as DatabaseQueryResult, e as DatabaseRow, f as DeepAgentActions, g as DockerActions, E as EmailActions, G as GitHubActions, I as IconifyActions, h as IntegrationActionName, i as IntegrationContracts, j as IntegrationName, L as LLMIntegrationActions, M as MLActions, O as OAuthActions, k as OtelActions, Q as QueueActions, R as RedisActions, S as StorageActions, l as StripeActions, T as TwilioActions, W as WikimediaActions, Y as YouTubeActions } from './contracts-CYtxZXrR.js';
+export { A as ArxivActions, C as CLIActions, D as DatabaseActions, a as DatabaseDriver, b as DatabaseQueryParamValue, c as DatabaseQueryParams, d as DatabaseQueryResult, e as DatabaseRow, f as DeepAgentActions, g as DockerActions, E as EmailActions, G as GitHubActions, I as IconifyActions, h as IntegrationActionName, i as IntegrationContracts, j as IntegrationName, L as LLMIntegrationActions, M as MLActions, O as OAuthActions, k as OtelActions, Q as QueueActions, R as RedisActions, S as StorageActions, l as StripeActions, T as TwilioActions, W as WikimediaActions, Y as YouTubeActions } from './contracts-CLTh6gjT.js';
 import { LogMeta } from '@almadar/core';
 export { I as IntegrationFactory, g as getIntegrationFactory, r as resetIntegrationFactory } from './factory-DTdVeyAi.js';
 export { GitHubIntegration } from './integrations/github/index.js';
@@ -344,14 +344,16 @@ declare class EmailIntegration extends BaseIntegration {
 /**
  * LLM integration using @almadar/llm
  *
- * Provides 4 actions:
+ * Provides 5 actions:
  * - generate: Generate text content from a prompt
  * - classify: Classify text into predefined categories
  * - extract: Extract structured data from text using a schema
  * - summarize: Summarize long text content
+ * - embed: Generate embeddings for an array of texts
  */
 declare class LLMIntegration extends BaseIntegration {
     private client;
+    private embeddings;
     private provider;
     constructor(config: IntegrationConfig);
     /**
@@ -359,11 +361,20 @@ declare class LLMIntegration extends BaseIntegration {
      * The client will throw a clear error when actually used without a key.
      */
     private getClient;
+    /**
+     * Lazily create the embeddings client. Embeddings route through a dedicated
+     * embedding model (default OpenRouter `baai/bge-base-en-v1.5`, matching
+     * `@almadar/curation-core`), independent of the chat provider/key, because
+     * most chat providers do not expose an embeddings endpoint. Override via
+     * `EMBEDDING_PROVIDER` / `EMBEDDING_MODEL` / `EMBEDDING_API_KEY`.
+     */
+    private getEmbeddings;
     execute(action: string, params: IntegrationParams): Promise<IntegrationResult>;
     private generate;
     private classify;
     private extract;
     private summarize;
+    private embed;
 }
 
 declare class MLIntegration extends BaseIntegration {

package/dist/index.js

@@ -5,7 +5,7 @@ import { google } from 'googleapis';
 import twilio from 'twilio';
 import sgMail from '@sendgrid/mail';
 import { Resend } from 'resend';
-import { getAvailableProvider, LLMClient } from '@almadar/llm';
+import { getAvailableProvider, LLMClient, EmbeddingClient } from '@almadar/llm';
 import { z } from 'zod';
 import { execSync, spawn } from 'child_process';
 import { mkdtempSync, writeFileSync, rmSync, promises } from 'fs';
@@ -997,6 +997,7 @@ var LLMIntegration = class extends BaseIntegration {
   constructor(config) {
     super(config);
     this.client = null;
+    this.embeddings = null;
     const configuredProvider = config.env.PROVIDER;
     this.provider = configuredProvider || getAvailableProvider() || "anthropic";
     this.logger.info(`LLM integration initialized (provider: ${this.provider})`);
@@ -1015,6 +1016,23 @@ var LLMIntegration = class extends BaseIntegration {
     }
     return this.client;
   }
+  /**
+   * Lazily create the embeddings client. Embeddings route through a dedicated
+   * embedding model (default OpenRouter `baai/bge-base-en-v1.5`, matching
+   * `@almadar/curation-core`), independent of the chat provider/key, because
+   * most chat providers do not expose an embeddings endpoint. Override via
+   * `EMBEDDING_PROVIDER` / `EMBEDDING_MODEL` / `EMBEDDING_API_KEY`.
+   */
+  getEmbeddings() {
+    if (!this.embeddings) {
+      this.embeddings = new EmbeddingClient({
+        provider: this.config.env.EMBEDDING_PROVIDER ?? "openrouter",
+        model: this.config.env.EMBEDDING_MODEL ?? "baai/bge-base-en-v1.5",
+        apiKey: this.config.env.EMBEDDING_API_KEY
+      });
+    }
+    return this.embeddings;
+  }
   async execute(action, params) {
     const validation = this.validateParams(action, params);
     if (!validation.valid) {
@@ -1045,6 +1063,9 @@ var LLMIntegration = class extends BaseIntegration {
         case "summarize":
           data = await this.executeWithRetry(() => this.summarize(params));
           break;
+        case "embed":
+          data = await this.executeWithRetry(() => this.embed(params));
+          break;
         default:
           throw new Error(`Unknown action: ${action}`);
       }
@@ -1137,6 +1158,15 @@ Return ONLY valid JSON matching the schema.`,
     });
     return result;
   }
+  async embed(params) {
+    const { texts } = params;
+    if (!Array.isArray(texts)) {
+      throw new Error("llm.embed: `texts` must be an array of strings");
+    }
+    this.logger.debug("Embedding texts", { count: texts.length });
+    const batch = await this.getEmbeddings().embedBatch(texts);
+    return { embeddings: batch.embeddings };
+  }
 };
 registerIntegration("llm", LLMIntegration);