Skip to main content
Getting an application running on Fly.io is essentially working out how to package it as a deployable image. Once packaged, it can be deployed to the Fly.io platform. In this guide we’ll learn how to deploy a FastAPI application on Fly.io. FastAPI is a modern, fast (high-performance), web framework for building APIs with Python based on standard Python type hints. Deploying a FastAPI app on Fly.io is… well it’s fast! You can be up and running in less than a minute.

​
Speedrun

First, install flyctl, the Fly.io CLI, and sign up to Fly.io if you haven’t already. The fastest way to get a basic Fast Api server on Fly.io is to use our FastAPI template:

​
Deploy a FastAPI app from scratch

For managing our project, we use Poetry. For more information on the initial setup with poetry, refer to setting up a python environment. We can initialize a new project like so:
If you’re using Poetry 2.0+ (the default when installed via pipx), the poetry shell command is no longer built in. Use poetry run to run commands (e.g., poetry run python main.py), or activate the environment with poetry env activate. See the initial setup guide for more details.
Then we have to add the FastAPI dependency:
Now, let’s create a simple FastAPI app in main.py:
We can then serve the development version of the app using the fastapi cli tool:
This will display a ‘hello from fly.io!’ message when you visit the root URL. With the development environment we also get cool features like hot-reloading. To run in production mode, we can amend the command:
The fastapi assumes your app is stored in main.py or app.py - if this is not the case, you must specify that when running the app:fastapi run hello/entrypoint.py
And with that you can deploy the app!
This will generate a fly.toml file with the configuration for your app and a Dockerfile that uses multi-stage builds. Refer to the fly.toml docs for more configuration options. To deploy a new version of your app, simply run fly deploy in the project directory. You can check out the full (yet minimal) example in this GitHub repository for a reference.