h2ogpte

H2OGPTe NodeJS/Typescript client

This client is experimental and subject to change.

Used for connecting to a Generative AI platform called H2OGPTe. See more at h2o.ai.

Installation

npm install h2ogpte

Auth

This client can authenticate in 2 ways.

• Using H2OGPTE API token (h2ogpte.setup('<endpoint_url>', '<token>')). See <h2ogpteinstance/api>.
• Using Azure Active Directory token (h2ogpte.setup('<endpoint_url>', '', '', '<azure_token>'))

Usage

import * as h2ogpte from 'h2ogpte'

async function main() {
  // Setup endpoint URL and access token.
  h2ogpte.setup(
    '<endpoint_url>',
    '<token>',
    '<optional CSRF token>',
    '<optional Active Directory token>'
  )

  // Create new collection with English as embedding model.
  const collectionID = await h2ogpte.createCollection(
    'Name',
    'Description',
    'English (bge-large-en-v1.5)'
  )

  // Create dummy File. In real world, it would come from filepicker (browser).
  const blob = new Blob(['This is some text content'], { type: 'text/plain' })

  // Upload the file
  const uploadID = await h2ogpte.uploadFile(
    new File([blob], 'myTextFile.txt', { type: 'text/plain' })
  )

  // Ingest the file.
  await h2ogpte.ingestUploads(
    'Ingesting documents',
    collectionID,
    [uploadID],
    false, // Generate document summaries.
    false // Generate document questions.
  )

  // Create a chat session.
  const { chatID } = await h2ogpte.createChatSession(collectionID)

  const onMessage = message => console.log(message)
  // Prepare a websocket client to start chatting. Pass onMessage callback.
  const socket = h2ogpte.Socket(onMessage)
  // Connect to server.
  const close = socket.connect()

  // Send a message.
  socket.send({
    t: 'cq',
    mode: 's',
    session_id: chatID,
    correlation_id: correlationID,
    body: 'Hello world',
    system_prompt: null,
    pre_prompt_query: null,
    prompt_query: null,
    pre_prompt_summary: null,
    prompt_summary: null,
    llm: null,
    llm_args: null,
    self_reflection_config: null,
    rag_config: null,
  })
  // Close the connection.
  close()

  // Fetch first 100 documents in a collection.
  const docs = await h2ogpte.listDocumentsInCollection(collectionID, 0, 99)
  // Summarize all documents in collection.
  const summaries = await Promise.all(
    docs.map(({ id }) => {
      return h2ogpte.createDocumentSummary(
        'Document summary',
        null,
        id,
        null,
        null,
        null,
        null,
        null,
        null,
        null,
        null
      )
    })
  )
}

main()

Underlying REST API client

The client utilizes a generated REST API client for establishing communication with the H2OGPTe server. This generated layer is also accessible and documented here.