PraisonAI - knowledge option in NodeJS/TypeScript library

[mits87]

I’m working with the PraisonAI Node.js/TypeScript library and want to use the knowledge option.
Compare to the python version for NodeJS/TypeScript version I can't find that option.

Anyone run it in TS?

Besides that I would like to:

• Take a file content as a Buffer (e.g., a PDF).
• Upload it as a knowledge source.
• Ensure that the vector DB / file is created only for the request and then removed automatically after processing.

Basically:

• I don’t want persistent storage.
• I’d like to have a “temporary knowledge base” attached to a single query.

Any idea?

[KeepALifeUS]

Knowledge support in the NodeJS/TypeScript library is valuable. Here is how to implement it:

Knowledge Layer Architecture

import { ChromaClient, Collection } from "chromadb";

interface KnowledgeConfig {
  collectionName: string;
  embeddingModel?: string;
  chunkSize?: number;
  chunkOverlap?: number;
}

interface KnowledgeDocument {
  id: string;
  content: string;
  metadata?: Record<string, any>;
}

class Knowledge {
  private client: ChromaClient;
  private collection: Collection | null = null;
  private config: KnowledgeConfig;

  constructor(config: KnowledgeConfig) {
    this.config = {
      chunkSize: 500,
      chunkOverlap: 50,
      ...config
    };
    this.client = new ChromaClient();
  }

  async initialize(): Promise<void> {
    this.collection = await this.client.getOrCreateCollection({
      name: this.config.collectionName
    });
  }

  async addDocument(doc: KnowledgeDocument): Promise<void> {
    const chunks = this.chunkText(doc.content);
    
    await this.collection?.add({
      ids: chunks.map((_, i) => `${doc.id}_chunk_${i}`),
      documents: chunks,
      metadatas: chunks.map(() => ({
        source_id: doc.id,
        ...doc.metadata
      }))
    });
  }

  async query(queryText: string, nResults = 5): Promise<string[]> {
    const results = await this.collection?.query({
      queryTexts: [queryText],
      nResults
    });
    
    return results?.documents?.[0] || [];
  }

  private chunkText(text: string): string[] {
    const chunks: string[] = [];
    const { chunkSize, chunkOverlap } = this.config;
    
    for (let i = 0; i < text.length; i += chunkSize! - chunkOverlap!) {
      chunks.push(text.slice(i, i + chunkSize!));
    }
    
    return chunks;
  }
}

Integration with PraisonAI Agent

import { Agent } from "praisonai";

const knowledge = new Knowledge({ collectionName: "docs" });
await knowledge.initialize();
await knowledge.addDocument({
  id: "doc1",
  content: "Your document content..."
});

const agent = new Agent({
  name: "KnowledgeAgent",
  instructions: "You have access to a knowledge base.",
  tools: [
    {
      name: "search_knowledge",
      description: "Search the knowledge base",
      execute: async (query: string) => {
        const results = await knowledge.query(query);
        return results.join("\n\n");
      }
    }
  ]
});

Key Features to Add

1. File loaders: PDF, Markdown, HTML
2. Auto-chunking: Smart sentence boundary detection
3. Metadata filtering: Query with filters
4. Caching: Avoid re-embedding unchanged docs

More patterns: https://github.com/KeepALifeUS/autonomous-agents