llmtk v3.5.7

LLM Toolkit

A toolkit of LLM related functions to simplify the dev process, especially with Vercel.

Chatbot

Setup

middleware.ts

import { middleware as serviceMiddleware } from 'llmtk/build/serverMiddleware';
import { NextRequestWithAuth, withAuth } from 'next-auth/middleware';
import { NextFetchEvent, NextResponse } from 'next/server';

async function middleware(request: NextRequestWithAuth, event: NextFetchEvent) {
  return (await serviceMiddleware(request)) || NextResponse.next();
}

export default withAuth(middleware);

api route

Create a file pages/api/google/tts.ts:

import { api } from 'llmtk/build/serverApi';

const { handler } = api.google.tts;
export default handler;

Usage

1. Use basic chat completion function on client side:

import { llm, services } from 'llmtk';

// Without stream
const reply = await llm.completion(services.openai.OpenAI_GPT3_5_4k, 'How are you');

// With stream
llm.completion(services.openai.OpenAI_GPT3_5_4k, 'How are you', {
  onStream: ({ acc }) => console.log(acc),
});

2. Use chat hook (include chat history and voice input/output):

import { llm, services } from 'llmtk';

const {
  // chat history
  history,

  // llm states
  replying,

  // text input
  sendTextMessage,
  // or voice input
  recorderStatus,
  toggleRecording,

  // voice output
  startSpeaking,
  stopSpeaking,
} = llm.useChat({
  model: services.openai.OpenAI_GPT3_5_4k,
  prompt: `You are C-3PO, human-cyborg relations.`,
  voiceLanguage: 'en-US',
  voiceName: 'en-US-Neural2-J',
});