cachedEmbeddings fix

packages/adapter-sqlite/src/index.ts

@@ -336,34 +336,29 @@ export class SqliteDatabaseAdapter extends DatabaseAdapter {
         query_field_name: string;
         query_field_sub_name: string;
         query_match_count: number;
-    }): Promise<
-        {
-            embedding: number[];
-            levenshtein_score: number;
-        }[]
-    > {
+    }): Promise<{ embedding: number[]; levenshtein_score: number }[]> {
         const sql = `
-      SELECT *
-      FROM memories
-      WHERE type = ?
-      AND vec_distance_L2(${opts.query_field_name}, ?) <= ?
-      ORDER BY vec_distance_L2(${opts.query_field_name}, ?) ASC
-      LIMIT ?
-    `;
-    console.log("sql", sql)
-    console.log("opts.query_input", opts.query_input)
-        const memories = this.db.prepare(sql).all(
+            SELECT 
+                embedding,
+                0 as levenshtein_score  -- Using 0 as placeholder score
+            FROM memories 
+            WHERE type = ?
+            AND json_extract(content, '$.' || ? || '.' || ?) IS NOT NULL
+            LIMIT ?
+        `;
+
+        const params = [
             opts.query_table_name,
-            new Float32Array(opts.query_input.split(",").map(Number)), // Convert string to Float32Array
-            opts.query_input,
-            new Float32Array(opts.query_input.split(",").map(Number))
-        ) as Memory[];
+            opts.query_field_name,
+            opts.query_field_sub_name,
+            opts.query_match_count
+        ];
 
-        return memories.map((memory) => ({
-            embedding: Array.from(
-                new Float32Array(memory.embedding as unknown as Buffer)
-            ), // Convert Buffer to number[]
-            levenshtein_score: 0,
+        const rows = this.db.prepare(sql).all(...params);
+
+        return rows.map((row) => ({
+            embedding: row.embedding,
+            levenshtein_score: 0
         }));
     }
 

packages/core/src/defaultCharacter.ts

@@ -4,9 +4,10 @@ export const defaultCharacter: Character = {
     name: "Eliza",
     plugins: [],
     clients: [],
-    modelProvider: ModelProviderName.LLAMALOCAL,
+    modelProvider: ModelProviderName.OPENAI,
     settings: {
-        secrets: {},
+        secrets: {
+        },
         voice: {
             model: "en_US-hfc_female-medium",
         },

packages/core/src/embedding.ts

@@ -4,7 +4,8 @@ import { fileURLToPath } from "url";
 import models from "./models.ts";
 import {
     IAgentRuntime,
-    ModelProviderName
+    ModelProviderName,
+    ModelClass
 } from "./types.ts";
 import fs from "fs";
 import { trimTokens } from "./generation.ts";
@@ -18,7 +19,7 @@ function getRootPath() {
     if (rootPath.includes("/eliza/")) {
         return rootPath.split("/eliza/")[0] + "/eliza/";
     }
-    
+
     return path.resolve(__dirname, "..");
 }
 
@@ -32,13 +33,13 @@ interface EmbeddingOptions {
 
 async function getRemoteEmbedding(input: string, options: EmbeddingOptions): Promise<number[]> {
     // Ensure endpoint ends with /v1 for OpenAI
-    const baseEndpoint = options.endpoint.endsWith('/v1') ? 
-        options.endpoint : 
+    const baseEndpoint = options.endpoint.endsWith('/v1') ?
+        options.endpoint :
         `${options.endpoint}${options.isOllama ? '/v1' : ''}`;
-    
+
     // Construct full URL
     const fullUrl = `${baseEndpoint}/embeddings`;
-    
+
     //console.log("Calling embedding API at:", fullUrl); // Debug log
 
     const requestOptions = {
@@ -87,7 +88,18 @@ async function getRemoteEmbedding(input: string, options: EmbeddingOptions): Pro
 export async function embed(runtime: IAgentRuntime, input: string) {
     const modelProvider = models[runtime.character.modelProvider];
     //need to have env override for this to select what to use for embedding if provider doesnt provide or using openai
-    const embeddingModel = modelProvider.model.embedding;
+    const embeddingModel = (
+        settings.USE_OPENAI_EMBEDDING ? "text-embedding-3-small" : // Use OpenAI if specified
+            modelProvider.model?.[ModelClass.EMBEDDING] || // Use provider's embedding model if available
+            models[ModelProviderName.OPENAI].model[ModelClass.EMBEDDING] // Fallback to OpenAI
+    );
+
+    if (!embeddingModel) {
+        throw new Error('No embedding model configured');
+    }
+
+    console.log("embeddingModel", embeddingModel);
+
 
     // Try local embedding first
     if (
@@ -107,16 +119,17 @@ export async function embed(runtime: IAgentRuntime, input: string) {
     // Get remote embedding
     return await getRemoteEmbedding(input, {
         model: embeddingModel,
-        endpoint: settings.USE_OPENAI_EMBEDDING ? 
+        endpoint: settings.USE_OPENAI_EMBEDDING ?
             'https://api.openai.com/v1' : // Always use OpenAI endpoint when USE_OPENAI_EMBEDDING is true
             (runtime.character.modelEndpointOverride || modelProvider.endpoint),
-            apiKey: settings.USE_OPENAI_EMBEDDING ? 
+        apiKey: settings.USE_OPENAI_EMBEDDING ?
             settings.OPENAI_API_KEY : // Use OpenAI key from settings when USE_OPENAI_EMBEDDING is true
             runtime.token,            // Use runtime token for other providers
         isOllama: runtime.character.modelProvider === ModelProviderName.OLLAMA && !settings.USE_OPENAI_EMBEDDING
     });
 }
 
+
 async function getLocalEmbedding(input: string): Promise<number[]> {
     const cacheDir = getRootPath() + "/cache/";
     if (!fs.existsSync(cacheDir)) {
@@ -137,13 +150,13 @@ export async function retrieveCachedEmbedding(
     runtime: IAgentRuntime,
     input: string
 ) {
-    if(!input) {
+    if (!input) {
         console.log("No input to retrieve cached embedding for");
         return null;
     }
-    
-    const similaritySearchResult = [];
-        // await runtime.messageManager.getCachedEmbeddings(input);
+
+    const similaritySearchResult =
+        await runtime.messageManager.getCachedEmbeddings(input);
     if (similaritySearchResult.length > 0) {
         return similaritySearchResult[0].embedding;
     }