Getting Started

Installation

zod-stream can be easily installed using your preferred package manager. Below are the commands for pnpm, npm, and bun.

With bun

$ bun add zod-stream zod openai

With pnpm

$ pnpm add zod-stream zod openai

With npm

$ npm install zod-stream zod openai

Basic Setup

To start using zod-stream in your project, follow these simple steps:

1. Define a response model

   Import and initialize the necessary modules from zod-stream, zod, and openai.

      import { z } from "zod";
    
      export const my_response_model = {
        schema: z.object({
          content: z.string()
        }),
        name: "bot response json"
      }

2. Create a structured LLM response using zod and zod-stream

   Ensure that your environment variables are properly configured, especially if you are interacting with external services like OpenAI.

     process.env["OPENAI_API_KEY"] = "your-openai-api-key"; // Replace with your OpenAI API key
     process.env["OPENAI_ORG_ID"] = "your-openai-org-id";   // Replace with your OpenAI Organization ID (if applicable)

   Import and initialize the necessary modules from zod-stream, zod, and openai, then pass your response model to the LLM provider using the withResponseModel utility.

      import { withResponseModel } from "zod-stream";
      import OpenAI from "openai";
    
      import { my_response_model } from "./my_response_model"
    
      const openAI = new OpenAI({
        apiKey: process.env["OPENAI_API_KEY"],
        organization: process.env["OPENAI_ORG_ID"]
      });
    
    
      // build completion params for the selected mode.
      const params = withResponseModel({
        response_model: my_response_model,
        params: {
          messages,
          model: "gpt-4"
        },
        mode: "TOOLS"
      })
      
      // make a completion call with your generated params
      const extractionStream = await oai.chat.completions.create({
        ...params,
        stream: true
      })
    
      
      // convert the default SSE response to a readable stream and then return as api response or pass elsewhere.
      const my_stream = OAIStream({
        res: extractionStream
      })

3. Read the partial streaming response from the stream

      import ZodStreamfrom "zod-stream";
      import OpenAI from "openai";
    
      import { my_response_model } from "./my_response_model"
    
      const zodstream = new ZodStream()
    
      const stream = await zodstream.create({
        completionPromise: async () => {
          // a promise that returns the stream created in step 2. This could be a fetch call or a promise that resolves the same way.
    
          const response = fetch("/api/get-stream", {
            body: JSON.stringify({ messages: [] }),
            method: "POST"
          })
    
          return response.body
        },
        // this schema should match the response model expected by the stream
        response_model: my_response_model
      })
    
      for await (const chunk of stream) {
        console.log(chunk) // safe to parse partial json
      }

This setup will prepare your application to handle structured LLM response streams effectively using zod-stream. For more advanced configurations and detailed usage, refer to the Usage and API Reference sections.