Built From Scratch with ❤️ by me | Video Demo

Nexus enables businesses to transform communication by boosting sales with intelligent AI assistants, securing calls from scammers, and offering 24/7 support via chatbots.

Nexusai was built from scratch with the following technologies:

• Gemini (Google AI Studio) (gemini-1.5-flash)
• Firebase (storage)
• Twilio (sms, phone number provisioning)
• Lemonsqueezy (subscription management)
• Puppeteer (web scraping)
• Cheerio (web scraping)
• Elevenlab (realistic text to speech)
• Browserless (offers headless browser automation)
• Cloudflare AI (qwen-1.5-14b-chat-awq)
• Telegram Bot API (telegram bot)
• NeonDB
• Postgresql
• Redis (caching)
• Upstash Qstash (bg-jobs)
• Nodejs
• React
• Tailwindcss
• Typescript
• Prisma (orm)

Headover to trynexusai.xyz to get started.

• Nodejs (v14 or higher) install
• Postgresql install
• Redis (for caching) install or brew (macos)
• Ngrok (for exposing localhost to the internet) install
• Twilio Account create
• Lemonsqueezy Account create

1. Clone the repo

git clone https://github.com/benrobo/nexusai.git

2. Install dependencies

# npm
cd nexusai && npm install

# yarn
cd nexusai && yarn

3. Create a .env file in the root directory and add all the environment variables in the .env.example file. Update the values accordingly.

4. Then run the following command to create the necessary tables and seed data.

# npm
npm run migrate

# yarn
yarn migrate

5. Using pgAdmin or any database client of your choice, headover to /packages/api/sql/function.sql copy and run the sql content inside the database sql editor.

-- Drop function
DROP FUNCTION IF EXISTS match_embeddings;

-- Create function
CREATE OR REPLACE FUNCTION match_embeddings (
  query_embedding text,
  match_threshold float,
  match_count int,
  kb_ids text[]
)
RETURNS TABLE (
  result json
)
LANGUAGE SQL STABLE
AS $$
  SELECT json_build_object(
    'id', public."knowledge_base_data".id,
    'similarity', 1 - (public."knowledge_base_data".embedding <=> query_embedding::vector),
    'content', string_to_array(public."knowledge_base_data".content, '.'),
    'metadata', public."knowledge_base_data".title
  ) AS result
  FROM
    public."knowledge_base_data"
  WHERE
    public."knowledge_base_data".kb_id = ANY(kb_ids) AND
    1 - (public."knowledge_base_data".embedding <=> query_embedding::vector) > match_threshold
  ORDER BY
    1 - (public."knowledge_base_data".embedding <=> query_embedding::vector) DESC
  LIMIT
    match_count;
$$;

That should create the necessary function for matching embeddings.

5. Update the Lemonsqueezy subscription plan in /packages/api/src/data/twilio/sub-plan.ts file, this values can be gotten from your lemonsqueezy store account.

First, headover to https://app.lemonsqueezy.com/products to create a product and variant for the subscription plan as seen in the image below.

Update your webhook URL in the Webhooks section of the product to <ngrok>.ngrok-free.app/api/webhook/tw-phone/subscription. Make sure you add in your signing secret in the .env file and select subscription_created and subscription_updated events as seen below.

Now, update the TwilioSubPlanInfo object in the /packages/api/src/data/twilio/sub-plan.ts file with the necessary values.

const TwilioSubPlanInfo = {
  local: {
    name: "Local",
    description: "Local phone numbers are specific to a city or region.",
    product: {
      test_id: 292219, // IN DEV MODE (update this)
      prod_id: 292219, // IN PROD MODE (update this)
    },
    variant: {
      test_id: 417672, // IN DEV MODE (update this)
      prod_id: 417672, // IN PROD MODE (update this)
    },
  },
} as TwilioSubPlanInfo;

The voice call feature is powered by Twilio, to try it out, you would need to do the following:

• Get a Twilio account here
• Upgrade the account.
• Get a starter ($5) Elevenlab account here
• Purchase a twilio phone number.
• Update the .env file
  • Make sure NODE_ENV is set to development otherwise, when purchasing a phone number from Nexus, you would be charged from your twilio balance (even if you've used a test credit card credentials).
  • Top up your account with atleast (min) $20.
• On purchasing a specific number from nexus, two records would be updated in the db: purchased_phone_numbers and used_phone_numbers. With the original twilio number you purchased directly with your twilio balance, copy that number and update the used_phone_numbers table with the phone fields.

Why? When you purchase a number on nexusai, your original balance ($20) isn't getting deducted from your twilio account since the number was purchased from nexusai in DEVELOPMENT_MODE as a result that number purchased isn't getting provisioned to your twilio account, which is why you first need to purchase the number directly from twilio. So, to make sure the number is linked to your account, you would need to update the used_phone_numbers table with the phone field with the number you purchased directly from twilio. your balance is only getting deducted when you either (1) puchase a number directly from twilio or (2) when you make a voice call or send sms.

• Headover to twilio and update the webhook URL for the phone number with <ngrok>.ngrok-free.app/api/webhook/tw-phone/voice. (You would only need to do this everytime your ngrok url get reset with a random new sub-domain).
• Now call that specific number linked to your agent.

Nexusai leverages firebase storage in caching audio files generated by the elevenlabs api. To get started, headover to the firebase console and create a new firebase project.

Create a new firebase storage bucket and update the FIREBASE.STORAGE_BUCKET in the .env file with the name of the bucket or better still, update necessary firebase config found in .env.example.

# FIREBASE
FIREBASE_API_KEY="xxxxxxx"
FIREBASE_AUTH_DOMAIN="<project-name>-<id>.firebaseapp.com"
FIREBASE_PROJECT_ID="<project-name>-<id>"
FIREBASE_STORAGE_BUCKET="<project-name>-<id>.appspot.com"
FIREBASE_MESSAGING_SENDER_ID="xxxxxx"
FIREBASE_APP_ID="1:xxxxx:web:xxxxxx"

5. Start the API server

# npm
npm run dev

# yarn
yarn dev

It should start the API server on http://localhost:4001 successfully.

1. Install dependencies

# npm
cd packages/app && npm install

# yarn
cd packages/app && yarn

2. Start the APP

# npm
npm run dev

# yarn
yarn dev

It should start the web app on http://localhost:4000 successfully.

1. Install dependencies

# npm
cd packages/chatbot && npm install

# yarn
cd packages/chatbot && yarn

2. Start the CHATBOT client

# npm
npm run dev

# yarn
yarn dev

It should start the CHATBOT client on http://localhost:3010 successfully.

The telegram bot server can be found in the /packages/tg-bot directory. To start the server, run the following commands:

# npm
cd packages/tg-bot && npm install

# yarn
cd packages/tg-bot && yarn

Run prisma generate to generate the prisma client.

Then create a .env file in the root directory and add the following environment variables:

NODE_ENV="development"

TG_BOT_TOKEN="xxxxxxxxxxxxxxxxxx"

API_URL="http://localhost:4001/api"

DATABASE_URL="postgres://postgres:12345@localhost:5432/nexusai"

Replace the TG_BOT_TOKEN with your telegram bot token gotten from the @BotFather.

Then start the server using the following command:

# npm
npm run dev

# yarn
yarn dev

It should start the telegram bot server on http://localhost:4005 successfully with the following message printed in the console:

2024-07-29 18:22:14 PM [info] : ✅ Telegram Bot launched successfully

In order to view the chatbot widget, start up the embed client located at /packages/embed/index.html using vscode-live-server or any other server of your choice.

• Update the CLIENT_URL and API_URL found in /nexus.js file with the local API_URL's i.e (Server) http://localhost:4001/api and (Chatbot)http://localhost:3010 respectively.

• Create a CHATBOT agent then Update the index.html script tag with your agent id.

<script src="./nexus.js" id="<your-agent-id>"></script>

After that, you should be able to view the chatbot widget on http://localhost:5500 (when started with vscode liver server extension only) successfully as seen below.

Considering the fact that this project was built within a short period of time (1.5 month to be exact), the codebase might not be as clean as expected. I would be working on refactoring the codebase and adding more features in the future (if time permits).

Response between production instance of nexusai might be slow, considering it was hosted on a cheap shared server instance on render with the following specs:

# server specs

- 1 CPU
- 2GB RAM

# database specs

- 1 vCPU
- 2GB RAM
- 4GB SSD

First Commit