Chatbots

Cover

Intro to LLMs in Xano

Cover

Build a Chatbot with ChatGPT & Xano

Building a Chatbot with OpenAI/ChatGPT and Xano

This guide will walk you through building a chatbot using ChatGPT and Xano.

Before we begin, it's important that you understand the following key concepts:

Database Basics

Building with Visual Development

APIs & Lambdas

User Authentication & User Data

You should know how to build a database table, build basic function stacks, work with user authentication, and utilize the External API Request function.

1

Understanding the OpenAI Chat Completions Endpoint

Objective: To create a chatbot, you'll primarily use OpenAI's chat completions API endpoint.

Endpoint: The specific endpoint is /v1/chat/completions. You'll make POST requests to this endpoint.

Authentication: All requests to the OpenAI API require authentication. This is done by including an Authorization header with a bearer token (your OpenAI API key).

Request Body:

  • model: Specifies which OpenAI model to use (e.g., gpt-3.5-turbo). You can find compatible models in the OpenAI documentation.

  • messages: This is a crucial part. It's an array containing the entire conversation history. Unlike interacting directly with ChatGPT, the API requires you to send all previous messages in each request.

  • Message Object Structure: Each object in the messages array needs a role and content:

    • role: Defines who sent the message.

      • system: Sets the initial context or persona for the chatbot (the first "training" prompt).

      • user: Represents messages sent by the end-user interacting with the bot.

      • assistant: Represents messages sent by the chatbot (responses from the API).

    • content: The actual text of the message.

Benefits of Sending Full History: This allows for fine-tuning or guiding the conversation by potentially modifying or constructing messages within the history you send to the API.

Other Parameters: There are optional parameters like temperature, but they aren't required to get started.

2

Define Database Schema

User Table: Create at least one test user for authentication purposes later.

Conversation Table (conversation): This acts as the parent table. Add the following fields:

  • name (Type: text): To easily identify conversations.

  • model (Type: text): To store which OpenAI model is used for this conversation (e.g., "gpt-3.5-turbo").

  • user_id (Type: Table Reference -> user): To link the conversation to the user who initiated it.

Messages Table (messages): This stores individual messages. Add the following fields, mirroring the structure needed for the OpenAI API request:

  • role (Type: text): Stores "system", "user", or "assistant".

  • content (Type: text): Stores the actual message text.

  • index (Type: integer): A number to track the order of messages within a conversation (0, 1, 2, ...). This is crucial for sorting messages correctly for display and for sending them in the right order to the API.

  • conversation_id (Type: Table Reference -> conversation): To link the message back to its parent conversation.

3

Create an endpoint to call OpenAI

API Group: Navigate to your API groups in Xano. You can use the default group or create a new one.

New API Endpoint: Add a new API endpoint. Choose "Start from scratch" or "Custom". Name it something descriptive, like send_message_to_openai.

Inputs: Define the necessary inputs for this endpoint. You'll likely need:

  • conversation_id (Input Type: Table Reference -> conversation): To know which conversation this message belongs to.

  • user_message (Input Type: text): The new message typed by the user.

Function Stack: This is where the logic happens using Xano's visual builder.

  1. Get OpenAI API Key: Securely retrieve your OpenAI API key. Store it in Xano's Environment Variables for security rather than hardcoding it.

  2. Get Conversation History:

    • Add a Query All Records step for the messages table.

    • Filter by the input conversation_id.

    • Sort by the index field in ascending order. This ensures messages are retrieved chronologically.

  3. Add User's New Message to History (Temporary): Add the incoming user_message to the list/array of messages retrieved in the previous step. Make sure it has the correct format: { "role": "user", "content": user_message }.

  4. Add External API Request: This is the core step to call OpenAI.

    • Click the "+" button in the function stack and select "External API Request".

    • Import cURL: Use the OpenAI documentation's cURL example for the chat completions endpoint. Copy the cURL command and use Xano's "Import cURL" feature. This will pre-fill most settings.

    • URL: Should be https://api.openai.com/v1/chat/completions.

    • Method: POST.

    • Headers:

      • Ensure Content-Type is application/json.

      • Add the Authorization header. The value should be Bearer YOUR_API_KEY, replacing YOUR_API_KEY dynamically using the environment variable retrieved in step 1. Use Xano's concatenation filters or sprintf for this.

    • Parameters/Body:

      • Set model to your desired model (e.g., "gpt-3.5-turbo"). You could make this dynamic based on the conversation record if needed.

      • Set messages to the full conversation history array you prepared in step 3 (including the new user message).

  5. Process API Response: The response from OpenAI will contain the assistant's reply. You'll typically find it in response.result.choices[0].message.content. Add steps to extract this content.

  6. Store Messages in Database:

    • Add a Add Record step for the messages table to save the user's new message. Include conversation_id, role ("user"), content (user_message), and the next index number.

    • Add another Add Record step for the messages table to save the assistant's response. Include conversation_id, role ("assistant"), content (the extracted response), and the subsequent index number.

  7. Response: Define what the Xano API endpoint should return to your frontend (e.g., the assistant's message content, or the updated full conversation).

4

Calling from a Frontend

Call the Xano API endpoint (send_message_to_openai) from your frontend application whenever a user sends a message.

Pass the conversation_id and the user_message.

Display the conversation history, potentially fetching it separately using the auto-generated Xano CRUD endpoints for the messages table, ensuring you sort by the index field.

Last updated

Was this helpful?