fly secrets become project secrets bound into the app. Fly is the easiest platform to leave, because a Fly app already has the one thing the others lack: a Dockerfile.
InstaCloud offers:
- Postgres with real branching. A branch opens as a copy-on-write clone holding all of the parent’s data, at its own connection string, so a feature or an agent task can run migrations nowhere near production.
- Built for coding agents. A CLI and an MCP server your agent drives directly, with an approval gate on the actions that matter.
- Flexible deployments. A local directory, a GitHub repo, or an image from any registry.
- Managed services alongside. Postgres, Redis, MySQL, MongoDB, and object storage in one project.
- Real support. An active community on Discord.
Migration steps
We will migrate Fly’s own Django example, a Python app with a Dockerfile. Its committedfly.toml, abridged, already holds most of what the migration needs:
1. Point a coding agent at your Fly app
Work in the app’s own checkout: theDockerfile and fly.toml it already holds are most of what the migration needs. Give your agent the skill there:
insta CLI, the insta skill, and the MCP server for every coding agent on this machine. The skill is what knows how a Fly app maps onto services here.
2. Create a project
3. Ask your agent to migrate
One thing first, because your agent cannot do it. Fly does not let anyone read a secret back:fly secrets list prints names and digests only. Either copy each value from wherever you set it, or read them off the running machine before you stop it:
DATABASE_URL that fly pg attach installed: your agent binds the new database under that name, and a copied string there would block the binding. Then hand over:
Prompt
insta deploy ., no GitHub connection needed. It will ask you to approve anything that touches production, and expect it to scale the Fly app to zero before it copies the database, so nothing is still writing.
Leave the Fly app stopped rather than destroyed, so you can go back to it. A machine left running would keep serving clients and writing to the old database.