gpt-preview README

A Node.js library and dev tool for previewing the output of an LLM or AI Agent (using an LLM)

• take complex output such as Function Calls or GraphQL mutations, and generate easy to view 'previews'

• can be used as a library in an AI application, to provide user with a preview of the AI actions, before the user actually agrees to apply the actions.

• can be used as a dev tool, by taking the AI output and pasting it into a locally run website.

• currently can summarize to:

  • DOT (graphviz) format
  • JSON format (containing short and long text summaries)

• supports LLM hosted on AWS Bedrock or OpenaI

Setup

• Node v20.18+

Usage [as command line tool]

• Clone this git repository

• Install dependencies

npm install

• Configure LLM via the config file:
  • AWS Bedrock config.
  • OpenAI config.

The config file needs to be named 'config.gpt-preview.json' and placed in the current working directory.

Then you can run the tool:

./go.sh <path to text file containing the LLM output> [OPTIONS]

- where OPTIONS are:
    -f=<DOT|JSON> (default is DOT)
    -o=<path to output file>

Usage [as an npm library]

npm install gpt-preview

To summarize text in-memory (to a variable 'summary') => JSON format:

import { OutputFormat, Platform, summarizeText } from "gpt-preview";

const config = {
  "platform": Platform.AwsBedrock,
  "isDebug": false,
  "modelId": "eu.anthropic.claude-3-5-sonnet-20240620-v1:0",
  "maxTokens": 2048,
  "temperature": 0.7,
  "top_p": 0.9,
  "awsRegion": "eu-west-1"
}

const summary = await summarizeText(
    "My LLM output to summarize",
    OutputFormat.JSON,
    config
    );

To summarize a file => DOT format:

import { OutputFormat, Platform, summarizeFile } from "gpt-preview";

const config = {
  "platform": Platform.AwsBedrock,
  "isDebug": false,
  "modelId": "eu.anthropic.claude-3-5-sonnet-20240620-v1:0",
  "maxTokens": 2048,
  "temperature": 0.7,
  "top_p": 0.9,
  "awsRegion": "eu-west-1"
}

const summary = await summarizeFile(
    "./temp/my-LLM-output.txt",
    OutputFormat.DOT,
    config,
    "./temp/my-summary.dot"
    );

To use OpenAI instead of AWS Bedrock, simply change the config:

const config = {
  "platform": Platform.OpenAI,
  "isDebug": false,
  "modelId": "gpt-4o-mini",
  "maxTokens": 2048,
  "temperature": 0.7,
  "top_p": 0.9
}