Everything GitHub Actions does for the spec also runs in one container on a machine you own. Fill in an .env file, start it, and it works on a schedule: it pulls the data every day, pushes what you merge, and commits every result back to your repo.
When a VM is the better fit
No GitHub Actions
Your code lives on GitLab, Bitbucket or your own git server, or Actions minutes are tight.
Several apps
One small machine runs the routines for every app, each with its own env file.
The weekly review on your Claude plan
Run the co-founder review with your Pro or Max subscription instead of an API key.
Your own keys stay on your machine
The store, Claude, fal, model and Search Console keys live in one .env file on a server you control.
Both can live side by side: the pull request check stays in Actions if you have it, and the VM does the scheduled work. Pick one place for push, so a merge is pushed once.
Pick a machine
Any Linux machine with Docker works. The routines clone the repo, run for a minute and exit, so the smallest plan of most providers is enough. The weekly review runs a coding agent, so give that one a little more memory.
- Hetzner
- DigitalOcean
- AWS
- Google Cloud
- Railway
- Fly.io
- Render
- Any Docker host
Nothing listens on a port. The container only calls GitHub, Appeeky, the stores and, if you set them, Claude and fal.
A token for one repo
On GitHub, create a fine-grained personal access token with Only select repositories → the app's repo, and Contents: Read and write. For the weekly review add Pull requests: Read and write. With any other git host, use a token that can push to that one repo.
# The app's repo, and a fine-grained token for it (Contents: read and write;# Pull requests: read and write too for the weekly review)REPO_URL=https://github.com/you/sleep-sounds.gitGITHUB_TOKEN=github_pat_…APPEEKY_API_KEY=apk_…# OptionalDAILY_AT=0 6 * * * # UTCPUSH_EVERY=*/15 * * * * # "off" to push by handWEEKLY_AT=0 7 * * 1 # the co-founder review, with the agent imageCLAUDE_CODE_OAUTH_TOKEN=… # claude setup-token: runs on your Claude planFAL_KEY=… # UGC clips on your own fal accountGSC_CREDENTIALS=/keys/gsc.json # Google Search Console: a service account keyASA_CLIENT_ID=SEARCHADS.… # Apple Search Ads API user (also ASA_TEAM_ID, ASA_KEY_ID,ASA_PRIVATE_KEY=/keys/asa.pem # ASA_PRIVATE_KEY): spend, installs and search termsVISIBILITY_EVERY=7 # ask the AI assistants every 7 days, on these keys:OPENAI_API_KEY=… # ChatGPT (also GEMINI_API_KEY, PERPLEXITY_API_KEY)HEALTHCHECK_URL=https://hc-ping.com/…PUSH_DRY_RUN=1 # the first week: show what a push would do
Start it
services:appeeky-spec:image: ghcr.io/appeeky/appeeky-spec:latest # or :latest-agentcommand: scheduleenv_file: .envrestart: unless-stopped
$ ssh root@your-vm$ curl -fsSL https://get.docker.com | sh$ mkdir appeeky-spec && cd appeeky-spec$ # put docker-compose.yml and .env here$ docker compose up -d$ docker compose logs -f
The first lines of the log show the schedule and where push starts following main:
$ docker compose logs -fRunning on this schedule (UTC):0 6 * * * daily*/15 * * * * push0 7 * * 1 weekly[push] Started at 3f2a91c. Merged listing changes are pushed from now on.
The schedule mode keeps running and restarts with the machine. To run one routine by hand: docker compose run --rm appeeky-spec daily.
What runs, and when
daily06:00 UTC
Category rank, AI visibility, Google search, reviews, competitors, social posts and UGC clips. Releases measured on day 7 and 14, verdicts written to the playbook. All committed to the repo.
pushevery 15 minutes
When main has a merged listing change, pushes it to the stores and records the release. Approved social drafts are sent and UGC briefs are made. A change the store cannot take yet waits for the next run.
weeklyMonday, optional
The co-founder review: six specialists read the repo, one change is proposed for a store that is free, with a brief, as a pull request. Needs the agent image and a Claude token.
Every routine starts from a fresh clone and ends with a commit, so the container keeps nothing between runs. Where push left off is a tag in the repo, appeeky-spec/pushed: replace the machine and it carries on from there.
A safe first week
Nothing old goes out
The first run starts from what is on main today and never pushes it. Only merges after that are pushed.
Dry run first
With PUSH_DRY_RUN=1, push prints what it would change and the posts and clips it would send, and changes nothing.
Know when it fails
HEALTHCHECK_URL is pinged after every daily run and gets /fail when it fails, so healthchecks.io or a similar service emails you.
Logs stay small
docker-compose.yml rotates them at 10 MB.
Container platforms and other schedulers
On Railway, Fly.io or Render, deploy the same image as a worker with no public port, set the variables from the .env file, and use schedule as the command. It behaves exactly like the VM.
Already have a scheduler, like cron, a Kubernetes CronJob or a CI you use elsewhere? Run one routine per job with MODE=daily, MODE=push or MODE=weekly. Each run does its work, commits and exits.
$ docker run --rm --env-file .env -e MODE=daily ghcr.io/appeeky/appeeky-spec:latest