Skip to content

Latest commit

 

History

History
84 lines (52 loc) · 3.64 KB

README.md

File metadata and controls

84 lines (52 loc) · 3.64 KB
Raw

Hatchet FastAPI Example

This is an example project demonstrating how to use Hatchet with FastAPI.

Prerequisites

Before running this project, make sure you have the following:

  1. Python 3.10 or higher installed on your machine.
  2. Poetry package manager installed. You can install it by running pip install poetry, or by following instructions in the Poetry Docs
  3. (Optional) If you would like to run the example frontend, Node which can be installed from the node website

Setup

  1. Create a .env file in the ./backend directory and set the required environment variables. This requires the HATCHET_CLIENT_TOKEN variable created in the Getting Started README. You will also need, a OPENAI_API_KEY which can be created on the OpenAI Website.

If you're running Hatchet locally without TLS:

HATCHET_CLIENT_TLS_STRATEGY=none
HATCHET_CLIENT_TOKEN="<token>"
OPENAI_API_KEY="<openai-key>"

If you're using Hatchet Cloud:

HATCHET_CLIENT_TOKEN="<token>"
OPENAI_API_KEY="<openai-key>"

  1. Open a terminal and navigate to the project backend directory (/fast-api-react/backend).

  2. Run the following command to install the project dependencies:

poetry install

Running the API

To start the FastAPI server, run the following command in the terminal:

poetry run api

Running the Hatchet Worker

In a separate terminal, start the the Hatchet worker by running the following command:

poetry run hatchet

(Optional) Running the Example Frontend Application

We've included a basic chat engine frontend to play with the example workflow. To run this script:

  1. Open a new terminal window and cd into the fast-api-react/frontend directory.
  2. run npm install
  3. run npm start
  4. By default you can access the application in your browser at http://localhost:3000 or by following the instructions in the terminal window.

Project Overview

Example Workflows

The project contains two example workflows in the ./backend/src/workflows directory. These workflows are registered with hatchet in ./backend/src/workflows/main.py which is started when running poetry run hatchet.

  1. Simple Response Generation: a single step workflow making a request to OpenAI
  2. Basic Retrieval Augmented Generation: a multi-step workflow to load the contents of a website with Beautiful soup, reason about the information, and generate a response with OpenAI.

Exposing the workflows via a RestAPI

A common design pattern is to start a Hatchet workflow run from a rest api endpoint. In this way, you can handle authentication and authorization as you normally do and let Hatchet handle execution. The simple FastAPI example can be found at ./backend/src/api/main.py

Starting a Run

The POST /message endpoint initiates a Hatchet workflow run, utilizing the message body as input. Given Hatchet operates asynchronously, this endpoint immediately returns a run ID. This ID acts as a reference for clients to track the status and outcomes of the initiated run.

Streaming Responses

After initiating a workflow run and receiving a run ID, clients can subscribe to updates through a GET /message/{id} request. This request allows clients to receive real-time notifications and results from the asynchronous Hatchet worker, associated with their specific run ID.