Chatterbox is a robust TypeScript library designed to work hand-in-hand between Swagger/OpenAPI documentation and OpenAI Function Calling. It allows us to tag endpoints in our Swagger documentation and automatically map them to function calls in OpenAI. This allows us to create a chatbot that can automatically call our API endpoints.
Chatterbox also handles the parsing of the generated response from OpenAI back into useful arguments or even a full API call.
npm install chatterbox-ai
# Or
bun add chatterbox-ai
- 📃 Automatically maps Swagger documentation to function calls.
- 💼 Supports both online fetching and local loading of Swagger documents.
- 🤖 Handles chat messages and converts them into back API calls.
To use Chatterbox, you'll need to import the package and instantiate it with specific tags.
Here's a simple example:
import Chatterbox from "chatterbox-ai";
const tagNames = ["Chat"];
const chatterbox = new Chatterbox(tagNames);
You can either fetch a Swagger document from a URL or load it from a local JSON object.
Fetch Swagger Doc
const swaggerUrl = `http://localhost:3000/api-docs-json`;
await chatterbox.fetchDoc(swaggerUrl);
Load Swagger Doc Locally
const swaggerDoc = document; /* your swagger doc as JSON object */
await chatterbox.loadDoc(swaggerDoc);
We can use the functionCalls
in our OpenAI chat creation endpoint. We also have a defaultSystemPrompt
that is a complementary
prompt that fits well with API calls.
const result = await openai.chat.completions.create({
model: "gpt-4",
messages: [
{
role: "system",
content: chatterbox.defaultSystemPrompt,
},
{
role: "user",
content: "Can you create a new document for me please?",
},
],
functions: chatterbox.functionCalls,
});
When parsing the generated response from OpenAI, we have 2 options. We can parse the arguments and just get the payload, or we can parse it directly to an API call, which will populate all of the fields and parameters for us.
Parse to Payload
const { message } = result.choices[0];
if (!message.function_call) continue;
// Payload is the arguments for the function call
// Endpoint is the OpenAPI path object
// Method and Path are the endpoint's method and path as strings
const { endpoint, method, path, payload } = chatterbox.parseMessage(message);
// Pass the arguments to your functions
await createDocument(payload);
Parse to API Call
When we use parseMessageToRequest
:
- Path params are automatically populated
- The other parameters are populated as query params or body params depending on the method
const req = chatterbox.parseMessageToRequest(message);
const { data } = await axios(req);