Week i-am-alice#1

files/OpenAIService.ts

@@ -52,7 +52,6 @@ export class OpenAIService {
    * @throws Error if there's an issue creating the embedding.
    */
   async createEmbedding(input: string | string[]): Promise<number[]> {
-    console.log(input)
     try {
       const embedding = await this.openai.embeddings.create({
         model: "text-embedding-3-large",

files/app.ts

@@ -27,7 +27,7 @@ app.post('/api/chat', async (req, res) => {
     if (!latestUserMessage) {
       return res.status(400).json({ error: 'No user message provided' });
     }
-
+    console.log(latestUserMessage);
     // Create embedding for the latest user message
     const latestMessageEmbedding = await openaiService.createEmbedding(latestUserMessage.content);
 

files/context/embeddings/memory_metadata.jsonl

@@ -1 +1,2 @@
+{"uuid":"14126146-b076-4f6b-a39f-7497199ff59a","filename":"family_town.md"}
 {"uuid":"a2baf2ce-e01d-4010-beb1-c06213fdbff5","filename":"adam_krakow.md"}

files/context/embeddings/message_metadata.jsonl

@@ -2,3 +2,6 @@
 {"uuid":"8206de41-6cf6-49c5-aff6-acb318661a8a","role":"assistant","filename":"540cc66a-02f4-48c1-8c01-6a6188aebdf5.md"}
 {"uuid":"2105c1b1-8f81-499f-bbd9-c000318c94f5","role":"assistant","filename":"16de94d4-8364-4aac-b48c-da586d083340.md"}
 {"uuid":"143e8d92-73a1-4433-9fa5-6a3619200e5d","role":"assistant","filename":"837c899f-da41-4130-9e6c-ac2a839c42b0.md"}
+{"uuid":"eae21d13-9401-42cd-a049-8f51aa6c3b8e","role":"assistant","filename":"030ccbe5-98d3-4309-aebb-62e55732ec6f.md"}
+{"uuid":"42931379-090e-4c79-96d5-6302f2bec87e","role":"assistant","filename":"2b33b71d-3917-46b8-9aca-e68fbec0b329.md"}
+{"uuid":"7e7173b4-903a-4378-b4a8-857853c49a36","role":"assistant","filename":"06879172-c891-4aaf-9d63-ec2724c110b8.md"}

memory/AssistantService.ts

@@ -34,6 +34,7 @@ export class AssistantService {
 
     async extractQueries(messages: ChatCompletionMessageParam[], trace: LangfuseTraceClient): Promise<string[]> {
         const generation = this.langfuseService.createGeneration(trace, "Extract queries", messages);
+
         try {
             const thread: ChatCompletionMessageParam[] = [
                 { role: "system", content: extractSearchQueriesPrompt({memoryStructure, knowledge: defaultKnowledge}) },
@@ -287,7 +288,11 @@ export class AssistantService {
                 messages: messagesWithSystem
             }) as ChatCompletion;
 
-            this.langfuseService.finalizeGeneration(generation, completion.choices[0].message, completion.model, completion.usage);
+            this.langfuseService.finalizeGeneration(generation, completion.choices[0].message, completion.model, {
+                promptTokens: completion.usage?.prompt_tokens,
+                completionTokens: completion.usage?.completion_tokens,
+                totalTokens: completion.usage?.total_tokens
+            });
             return completion;
         } catch (error) {
             this.langfuseService.finalizeGeneration(generation, { error: error instanceof Error ? error.message : String(error) }, "unknown");

memory/MemoryService.ts

@@ -7,6 +7,7 @@ import type { OpenAIService } from './OpenAIService';
 import { VectorStore } from './VectorStore';
 import { LangfuseService } from './LangfuseService';
 import { LangfuseTraceClient } from 'langfuse';
+import { execSync } from 'child_process';
 
 export interface Memory {
   uuid: string;
@@ -95,7 +96,7 @@ export class MemoryService {
     // Add hashtags at the end of the file
     if (frontmatterData.metadata?.tags && frontmatterData.metadata.tags.length > 0) {
       markdownContent += '\n\n';
-      markdownContent += frontmatterData.metadata.tags.map(tag => `#${tag}`).join(' ');
+      markdownContent += frontmatterData.metadata.tags.map(tag => `#${tag.replace(/\s/g, '_')}`).join(' ');
     }
 
     return markdownContent;
@@ -111,7 +112,7 @@ export class MemoryService {
 
     // Ensure tags in metadata match those at the end of the file
     if (hashtags) {
-      const tagsFromContent = hashtags.split(' ').map(tag => tag.replace('#', ''));
+      const tagsFromContent = hashtags.split(' ').map(tag => tag.replace('#', '').replace(/_/g, ' '));
       data.metadata.tags = [...new Set([...(data.metadata.tags || []), ...tagsFromContent])];
     }
 
@@ -230,15 +231,19 @@ export class MemoryService {
     );
   }
 
-  async searchSimilarMemories(query: string, k: number = 15): Promise<Memory[]> {
+  async searchSimilarMemories(query: string, k: number = 15): Promise<Array<Memory & { similarity: number }>> {
     const queryEmbedding = await this.openaiService.createEmbedding(query);
-    const similarIds = await this.vectorStore.search(queryEmbedding, k);
-    if (similarIds.length === 0) {
+    const similarResults = await this.vectorStore.search(queryEmbedding, k);
+    if (similarResults.length === 0) {
       console.log('No similar memories found.');
       return [];
     }
-    const memories = await Promise.all(similarIds.map(id => this.getMemory(id)));
-    return memories.filter((m): m is Memory => m !== null);
+    const memories = await Promise.all(similarResults.map(result => this.getMemory(result.id)));
+    return memories.filter((m): m is Memory => m !== null)
+      .map((memory, index) => ({
+        ...memory,
+        similarity: similarResults[index].similarity
+      }));
   }
 
   async deleteMemory(uuid: string): Promise<boolean> {
@@ -309,4 +314,75 @@ export class MemoryService {
     const urls = memory.metadata?.urls && memory.metadata.urls.length > 0 ? `\nURLs: ${memory.metadata.urls.join(', ')}` : '';
     return `<memory uuid="${memory.uuid}" name="${memory.name}" category="${memory.category}" subcategory="${memory.subcategory}" lastmodified="${memory.updated_at}">${memory.content.text}${urls}</memory>`;
   }
+
+  async syncMemories(trace: LangfuseTraceClient): Promise<{ added: string[], modified: string[], deleted: string[] }> {
+    const gitDiff = this.getGitDiff();
+    const changes = this.parseGitDiff(gitDiff);
+
+    console.log(changes);
+    const added: string[] = [];
+    const modified: string[] = [];
+    const deleted: string[] = [];
+
+    for (const file of changes.added) {
+      const memory = await this.addMemoryFromFile(file);
+      if (memory) added.push(memory.uuid);
+    }
+
+    for (const file of changes.modified) {
+        console.log('Updating file', file)
+      const memory = await this.updateMemoryFromFile(file);
+      if (memory) modified.push(memory.uuid);
+    }
+
+    for (const file of changes.deleted) {
+      const success = await this.deleteMemoryByFile(file);
+      if (success) deleted.push(file);
+    }
+
+    this.langfuseService.createEvent(trace, "SyncMemories", { added, modified, deleted });
+    return { added, modified, deleted };
+  }
+
+  private getGitDiff(): string {
+    const command = 'git diff --name-status HEAD';
+    return execSync(command, { cwd: path.join(this.baseDir) }).toString();
+  }
+
+  private parseGitDiff(diff: string): { added: string[], modified: string[], deleted: string[] } {
+    const lines = diff.split('\n');
+    const changes = { added: [], modified: [], deleted: [] };
+
+    for (const line of lines) {
+      const [status, file] = line.split('\t');
+      if (!file || !file.endsWith('.md')) continue;
+
+      if (status === 'A') changes.added.push(file);
+      else if (status === 'M') changes.modified.push(file);
+      else if (status === 'D') changes.deleted.push(file);
+    }
+
+    return changes;
+  }
+
+  private async addMemoryFromFile(file: string): Promise<Memory | null> {
+    const filePath = path.join(this.baseDir, 'memories', file);
+    const content = await fs.readFile(filePath, 'utf-8');
+    const memory = this.markdownToJson(content);
+    return this.createMemory(memory, {} as LangfuseTraceClient); // Note: We need to handle the trace properly here
+  }
+
+  private async updateMemoryFromFile(file: string): Promise<Memory | null> {
+    const filePath = path.join(this.baseDir, 'memories', file);
+    const content = await fs.readFile(filePath, 'utf-8');
+    const updatedMemory = this.markdownToJson(content);
+    return this.updateMemory(updatedMemory, {} as LangfuseTraceClient); // Note: We need to handle the trace properly here
+  }
+
+  private async deleteMemoryByFile(file: string): Promise<boolean> {
+    const filePath = path.join(this.baseDir, 'memories', file);
+    const content = await fs.readFile(filePath, 'utf-8');
+    const memory = this.markdownToJson(content);
+    return this.deleteMemory(memory.uuid);
+  }
 }

memory/VectorStore.ts

@@ -40,16 +40,34 @@ export class VectorStore {
     }
   }
 
-  async search(vector: number[], k: number): Promise<string[]> {
+  async search(vector: number[], k: number): Promise<Array<{ id: string; similarity: number }>> {
     try {
       const normalizedVector = this.normalizeVector(vector);
       const totalVectors = this.index.ntotal();
       if (totalVectors === 0) {
         return [];
       }
       const actualK = Math.min(k, totalVectors);
-      const { labels } = this.index.search(normalizedVector, actualK);
-      return labels.map(label => this.metadata.get(label) || '');
+      const { distances, labels } = this.index.search(normalizedVector, actualK);
+
+      const results = labels.map((label, index) => ({
+        id: this.metadata.get(label) || '',
+        similarity: distances[index]
+      }));
+
+
+      console.log(`Total results: ${results.length}`);
+
+      // Calculate average similarity
+      const avgSimilarity = results.reduce((sum, r) => sum + r.similarity, 0) / results.length;
+
+      // Filter results with at least 80% of average similarity
+      const threshold = avgSimilarity * 0.8;
+      const filteredResults = results.filter(r => r.similarity >= threshold);
+
+      console.log(`Filtered results: ${filteredResults.length}`);
+
+      return filteredResults;
     } catch (error) {
       console.error('Error searching vectors:', error);
       return [];
@@ -80,13 +98,18 @@ export class VectorStore {
     }
   }
 
-
   update(embedding: number[], id: string): void {
-    const index = this.ids.indexOf(id);
-    if (index !== -1) {
-      this.embeddings[index] = embedding;
+    const normalizedEmbedding = this.normalizeVector(embedding);
+    const existingIndex = Array.from(this.metadata.entries()).find(([_, value]) => value === id)?.[0];
+    if (existingIndex !== undefined) {
+      // Remove the existing vector
+      this.index.removeIds([existingIndex]); // Changed from remove_ids to removeIds
+      // Add the new vector
+      this.index.add(normalizedEmbedding);
+      // Update the metadata
+      this.metadata.set(this.index.ntotal() - 1, id);
     } else {
-      this.add(embedding, id);
+      this.add(normalizedEmbedding, id);
     }
   }
 }

memory/app.ts

@@ -6,6 +6,8 @@ import { AssistantService } from './AssistantService';
 import { defaultKnowledge as knowledge } from './prompts';
 import { LangfuseService } from './LangfuseService';
 import type { ChatCompletionMessageParam } from "openai/resources/chat/completions";
+import { execSync } from 'child_process';
+import path from 'path';
 
 const app = express();
 const port = 3000;
@@ -39,6 +41,21 @@ app.post('/api/chat', async (req, res) => {
   }
 });
 
+app.post('/api/sync', async (req, res) => {
+  const trace = langfuseService.createTrace({ id: uuidv4(), name: 'Sync Memories', sessionId: uuidv4() });
+
+  try {
+    const changes = await memoryService.syncMemories(trace);
+    await langfuseService.finalizeTrace(trace, {}, changes);
+    await langfuseService.flushAsync();
+    return res.json(changes);
+  } catch (error) {
+    await langfuseService.finalizeTrace(trace, {}, { error: 'An error occurred while syncing memories' });
+    console.error('Error in memory synchronization:', error);
+    res.status(500).json({ error: 'An error occurred while syncing memories' });
+  }
+});
+
 app.listen(port, () => console.log(`Server running at http://localhost:${port}`));
 
 process.on('SIGINT', async () => {

memory/memories/index.jsonl

@@ -0,0 +1 @@
+{"category":"profiles","subcategory":"basic","name":"adam","content":{"text":"Adam Gospodarczyk is an entrepreneur, full-stack developer specializing in JS, Node.js, and Rust, and a designer. He is based in Krakow, Poland. Adam is known for his curiosity and as a lifelong learner. He is involved in several projects including eduweb.pl, heyalice.app, easy.tools, Tech•sistence, Ahoy!, overment, AI_devs, and Zautomatyzowani.pl."},"metadata":{"confidence":100,"urls":[],"tags":["adam gospodarcyk","entrepreneur","full-stack developer","designer","krakow","projects"]},"uuid":"8d9617cc-fa9d-4a24-9c2b-8773127dfb4c","created_at":"2024-09-14T12:25:32.253Z","updated_at":"2024-09-14T12:25:32.253Z"}

memory/memories/profiles/basic/adam.md

@@ -0,0 +1,24 @@
+---
+category: 'profiles'
+subcategory: 'basic'
+name: 'adam'
+metadata:
+  confidence: 100
+  urls: []
+  tags:
+    - 'adam gospodarcyk'
+    - 'entrepreneur'
+    - 'full-stack developer'
+    - 'designer'
+    - 'krakow'
+    - 'projects'
+uuid: '8d9617cc-fa9d-4a24-9c2b-8773127dfb4c'
+created_at: '2024-09-14T12:25:32.253Z'
+updated_at: '2024-09-14T12:25:32.253Z'
+---
+
+Adam Gospodarczyk is an entrepreneur, full-stack developer specializing in JS, Node.js, and Rust, and a designer. He is based in Krakow, Poland. Adam is known for his curiosity and as a lifelong learner. He is involved in several projects including eduweb.pl, heyalice.app, easy.tools, Tech•sistence, Ahoy!, overment, AI_devs, and Zautomatyzowani.pl.
+
+Adam is 193 cm tall.
+
+#adam_gospodarcyk #entrepreneur #full-stack_developer #designer #krakow #projects

memory/memories/vector_metadata.json

@@ -0,0 +1 @@
+[[0,"8d9617cc-fa9d-4a24-9c2b-8773127dfb4c"]]