Dockerfile and floo.app.toml. Map its processes and scheduled jobs, import its config vars, restore its database, and move its domain.
Requires the app’s code in a GitHub repository, the floo CLI signed in with floo auth login, and the Heroku CLI signed in to the app. Without a GitHub repository, floo deploy --hosted deploys from a copy floo hosts; see Deploy without a GitHub repository.
In its March 2026 update, Heroku said it has moved to a sustaining engineering model.
What maps to what
1. Add a Dockerfile
floo builds every service from aDockerfile. Start from a recommended Dockerfile or the stack guide for Rails, Django, FastAPI, Express, or Next.js. Install the same runtime version the buildpack used, and make the final stage start the web command from the Procfile.
Add a .dockerignore so local dependency folders and .env* files stay out of the build context.
2. Map the Procfile
A Python app with a worker, a release command, and a nightly Heroku Scheduler job:Procfile
floo init my-app, then edit floo.app.toml to match the processes and jobs:
floo.app.toml
command runs as sh -c, so $PORT expands. The worker shares the web service’s image, so it needs its own command. Add migrate_command = "alembic upgrade head" to [services.web] after the data is in (step 5). Set access_mode = "accounts" under [app] if only invited people should reach the app (App sign-in).
Each Heroku Scheduler job becomes a [cron.<name>] block with the job’s command. Scheduler’s three intervals are */10 * * * * (every 10 minutes), 0 * * * * (hourly), and a daily expression such as 0 3 * * *. floo reads schedules in UTC, runs a job for up to 1,500 seconds, and does not retry it.
3. Import config vars
heroku config --shell prints config vars in shell format. Exclude variables for Heroku’s add-ons; floo injects DATABASE_URL, PG*, and Redis URLs for its managed services. Add .env.heroku and .env.import to .gitignore before creating them:
.env.import after the first deploy creates the app. If the app reads REDIS_URL, point it at REDIS_URL_CACHE.
4. Deploy to dev
floo deploy stops with one link if floo can’t read the repository yet. Install the floo GitHub App on the repository’s owner, then run floo deploy again. It waits until the dev deploy is live, and floo status --app my-app --json shows why if it is not. The first build or startup can fail if it needs variables that have not been imported yet. Once the app exists, import them and rebuild:
.env.import until you have imported the prod values in step 6; then delete both files.
5. Move the database
The dev deploy creates the managed database. Capture a Heroku Postgres backup, download it, convert it to SQL, and load it into floo’s dev database withpsql:
pg_dump’s custom format, and Heroku’s own restore example names the file latest.dump. floo run sets the PG* variables for the dev database, so psql connects with no connection string. If the load stops on the heroku_ext schema, create it first, as Heroku’s docs advise:
migrate_command to [services.web] and push. The dump carries your migration tool’s history table, so the first run finds nothing to apply.
floo run opens a dev session, so these commands load dev. Prod gets its own database with the first release (step 6). After that release, load the same file into prod with prod’s DATABASE_URL:
floo env get needs a write-scoped floo API key, and each read is audited. --reveal-secrets lets it print a connection string, and the command substitution keeps the URL out of your shell history. --clean --if-exists in the converted file drops each table before recreating it, so the load also works after the release’s migrate_command has created the schema. To export a managed database, see floo db dump.
6. Move the domain
A custom domain serves prod. Replaceowner/repo with your GitHub repository and publish the first release:
.env.heroku and .env.import after verifying the prod import. floo domains show prints two records:
Publish the certificate record first; it does not change where traffic goes. Then replace the record that points at the Heroku DNS target (the
<name>.herokudns.com value heroku domains --app my-heroku-app lists, per Heroku’s custom domains article) with the floo one. For an apex such as example.com, use an ALIAS or ANAME to edge.getfloo.com, or an A record to 34.117.26.37.
active. Pointing the traffic record back at Heroku is the move’s rollback.
Roll back on floo
Choose a deploy ID from the list, check its environment, and replaceDEPLOY_ID:
heroku rollback restores the slug and its config vars. floo rollback re-points traffic at an earlier image only: env vars keep their current values, and migrate_command does not run. See Releases and rollbacks.
Review apps and pipelines
Heroku creates a review app for each pull request once review apps are on. floo previews are off until you turn them on; each pull request then gets its own URL and its own Postgres branch:partial
gh release create publishes that commit to prod. See Preview environments and Production environments.
Agent prompt
Paste into your agent from the repository root. Replacemy-app, my-heroku-app, and <owner/repo>.
Where Heroku is stronger
As of November 9, 2026, from Heroku’s documentation:- No Dockerfile. Buildpacks detect the language and build the app with minimal configuration. floo builds from a Dockerfile you keep in the repository.
- One-off dynos in the cloud.
heroku runstarts a dyno next to the app’s other dynos.floo runruns the command on your machine with the dev env vars. - Rollback includes config.
heroku rollbackrestores the build and its config vars together.floo rollbackrestores the image only. - Postgres extras. Standard-tier databases can add followers for read scaling, and Dataclips share query results as links. floo offers neither.
- Review apps by default. With review apps on, Heroku creates one as soon as a pull request opens. floo previews are opt-in per app or per pull request.
What it costs
One small full-stack app, priced as of November 9, 2026:- One web service with light traffic.
- A small Postgres database, about 1 GB.
- Sign-in so about 5 named people can use the app, and nobody else.
- One job that runs once a day.
- floo’s numbers are the personal-app estimate on getfloo.com/pricing: “a default-size service (1 vCPU, 0.5 GiB) busy about an hour a day, a 1 GiB database and light traffic. Builds and logs are extra.”
floo’s Pro plan is $25 a month with $25 of credits and 5 builder seats; Team is $250 a month with $250 of credits and 50 builder seats. Rates are on getfloo.com/pricing.
Next steps
Dockerfiles
Requirements, layer order, and recommended Dockerfiles.
Postgres
Injected variables, migrations, and preview branches.
Cron jobs
Schedules, timeouts, manual runs, and failure handling.
Custom domains
DNS records, apex domains, and activation.