Flyte Devbox in a Codespace

Try Flyte in your browser with zero local setup. This repo ships a devcontainer that automatically starts a Flyte devbox — a complete single-node Flyte cluster running inside the codespace — so you can get a feel for the platform in a few minutes.

Machine type: the 4-core machine is recommended — the devbox runs a Kubernetes cluster in Docker and appreciates the headroom — but the 2-core machine works too since the examples request only fractional CPUs.

What happens on startup

1. The devcontainer installs the flyte SDK and writes .flyte/config.yaml pointing at the devbox (localhost:30080, local image builder).
2. flyte start devbox boots the local cluster. The first boot pulls container images and takes a few minutes — watch progress in the "Running postStartCommand" terminal.
3. Port 30080 (the Flyte UI and API) is forwarded automatically over plain HTTP and switched to public visibility so the UI link just works. If the automatic switch fails (it needs the codespace scope on your GitHub token), right-click the port in the PORTS tab and set Port Visibility → Public.

Run your first workflow

Once the devbox is up:

flyte run examples/hello.py main

The CLI prints a link to the run. Open the PORTS tab in VS Code and click the forwarded address for port 30080 to browse the Flyte UI — you'll see the run's task graph, inputs/outputs, and logs.

Examples

All examples request tiny amounts of CPU (250m–500m) so they fit even on the smallest 2-core codespace:

Example	What it shows
examples/hello.py	The smallest workflow: one task calling another
examples/parallel_map.py	Fanning out parallel tasks with flyte.map
examples/pipeline.py	A typed extract → transform → summarize pipeline

flyte run examples/parallel_map.py main --n 20
flyte run examples/pipeline.py main --n 500

Add --local (e.g. flyte run --local examples/hello.py main) to run in your Python process instead of on the devbox cluster.

Handy commands

flyte get runs                 # list past runs
flyte stop devbox              # shut the cluster down
flyte start devbox             # bring it back
flyte delete devbox --volume   # remove it entirely, including data
kubectl get pods -A            # peek under the hood

Next steps

• Flyte documentation
• Run modes: local, devbox, remote
• Deploy Flyte to a real cluster