▋
packages/chat-orchestrator/README.md

@portfolio/chat-orchestrator

OpenAI integration and chat orchestration logic.

Overview

This package handles the core chat pipeline:

  • OpenAI API integration (GPT-4, embeddings)
  • Multi-stage RAG orchestration
  • Token counting and management
  • Streaming response generation

Dependencies

  • openai - OpenAI API client
  • js-tiktoken - Token counting
  • @portfolio/chat-contract - Type contracts
  • @portfolio/chat-data - Document retrieval

Architecture

Chat Pipeline

  1. Request Validation - Validate incoming message
  2. Query Planning - Determine retrieval strategy
  3. Document Retrieval - Fetch relevant context
  4. Response Generation - Stream LLM response

Models

Configured in chat.config.yml:

models:
  plannerModel: gpt-5-mini-2025-08-07
  answerModel: gpt-5-mini-2025-08-07
  embeddingModel: text-embedding-3-small
  reasoning:
    planner: low
    answer: low

Usage

import { createChatOrchestrator } from '@portfolio/chat-orchestrator';

const orchestrator = createChatOrchestrator({
  openaiApiKey: process.env.OPENAI_API_KEY,
  config: chatConfig,
});

// Stream response
for await (const chunk of orchestrator.chat(message, context)) {
  process.stdout.write(chunk);
}

Token Management

Uses js-tiktoken for accurate token counting:

tokens:
  planner: 15000
  answer: 15000

Related Packages