Fastify Server Template done in TypeScript.

Fatsy is using the same route definition standard as in Foxx Builder, Umi, Ice, and many others:

/routes
-- get.js
-- /echo
---- get.js
-- /users
---- /$id
------- get.js
---- get.js
---- post.js

The routes folder should contain all your routes. Each route is a folder with a name that is the same as the route path. The route path is the path that is used in the browser to access the route. For example, if you have a route /users, then the route path is /users. If your route has a parameter, then the route path contains a $ followed by the name of the parameter.

Each route folder can contain one or more files. Each file is a route handler. The name of the file is the HTTP method that the route handler handles. For example, if you have a route handler that handles the POST method, then the name of the file is post.js. If you have a route handler that handles the GET method, then the name of the file is get.js.

In order to support websockets, you need to create a file ws.js in the route folder. This file will be used as a websocket handler.

The following example demonstrates how to create a websocket handler that will use OpenAI to answer questions from the user. It is using Weaviate as a vector store for the document retrieval and index.

import weaviate from 'weaviate-ts-client';
import {WeaviateStore} from 'langchain/vectorstores/weaviate';
import {OpenAIEmbeddings} from 'langchain/embeddings/openai';

const ws: FatsyDynamicRoute = {
    private: false,
    handler: async (connection: SocketStream, _request: FastifyRequest) =>
    {
        const client = weaviate.client({
            scheme: process.env.WEAVIATE_SCHEME || 'http',
            host: process.env.WEAVIATE_HOST || 'localhost'
        });

        const store = await WeaviateStore.fromExistingIndex(new OpenAIEmbeddings(), {
            client,
            indexName: 'Test',
            textKey: 'content',
            metadataKeys: ['productId', 'versionId']
        });

        const chat = new OpenAI({
            streaming: true,
            callbacks: [
                {
                    handleLLMNewToken(token)
                    {
                        connection.socket.send(token);
                    }
                }
            ]
        });

        const chain = ConversationalRetrievalQAChain.fromLLM(
            chat,
            store.asRetriever()
        );

        connection.socket.on('message', async (message: string) =>
        {
            await chain.call({
                question: message.toString(),
                chat_history: []
            });
        });

        connection.socket.on('close', () =>
        {
            console.log('Client disconnected');
        });
    }
};

export default ws;

No more confusing JSON, - simple and readable YAML config

app:
  title: My demo API server

server:
  host: "0.0.0.0"
  port: 3000
  logger:
    level: trace

• Application routing convention similar to the one we use on Foxx Builder
• Application configuration with load order to the config npm package, so the first one loads is default, then local, then your NODE_ENV profile
• Templating engine - by default, there is Handlebars bundled for you
• S3 storage support (Example)
• Websockets support

• Runs in standalone mode
• Runs in Docker
• Runs in Amazon ECS
• Runs in amazon ECS via HashiCorp waypoint

• Connect Node.js to MinIO with TLS using AWS S3