AFK

A CLI to run AI development workloads inside your cloud environment. Launch a coding agent in your own AWS, GCP, or Cloudflare account and come back to the results.

Get started View on GitHub

# In any repo with an afk.Dockerfile:
afk init --provider aws                                    # point AFK at your cloud
afk golden build                                           # one-time: build the boot image
afk secrets put github-token <PAT>                         # so the Run can clone your repo
echo "GITHUB_TOKEN=secret:github-token" >> .afk.env
afk run "claude -p 'fix the failing test and open a PR'"   # launch an agent in your cloud
afk ls                                                     # see it running
afk logs <run-id>                                          # tail its output

That Run boots an ephemeral instance in your own AWS account, does the work, and terminates. The same commands work unchanged on GCP and Cloudflare.

Why AFK

One CLI to rule your clouds

The same afk run, afk attach, afk ls, afk kill on AWS, GCP, and Cloudflare. Switch providers without changing a command.

Ephemeral by design

Each Run boots, does its work, and self-terminates. No always-on infrastructure to maintain or pay for.

Your dev environment, ported

A Dockerfile and compose file describe the Run — so your entire dev and test environment (toolchain, Postgres, Redis, …) travels to the cloud and is available to the agent.

Stay in the loop

Tail an agent’s logs live, attach a shell into its container to steer it, or kill it — from wherever you are.

Pick a path

Introduction What AFK is, the model behind it, and how a Run flows end to end.

Installation Clone, bun link, and the per-Backend prerequisites.

Quickstart Stand up your first Run on AWS, GCP, or Cloudflare.

Glossary The canonical vocabulary — Run, Backend, Golden Image, Ref, and more.