Skip to main content
Version: Next

Upgrade to 3.5

Serverpod 3.5 (Jetstream) brings a unified serverpod start command with hot reload that runs your server, database, and Flutter app together. The release also includes an embedded Postgres option and optional AI agent skills for your editor. The changes are mostly opt-in: your existing 3.4 project keeps working with small updates.

This guide walks through the upgrade and should take about 15 minutes.

Before you start

  • Your project is on the latest Serverpod 3.4.x release.
  • Your project compiles and tests pass.
  • You've committed your current state to Git so you can roll back if needed.

Update the Serverpod CLI

Install the 3.5 CLI:

$ dart install serverpod_cli 3.5.0-beta.9

Verify the version:

$ serverpod version

Update your project dependencies

In each package's pubspec.yaml (<project>_server, <project>_client, <project>_flutter), bump the Serverpod packages to 3.5. Serverpod prefers an exact version pin over a caret range to keep the CLI and the packages in sync:

dependencies:
serverpod: 3.5.0-beta.9
serverpod_client: 3.5.0-beta.9 # in the client and Flutter packages
serverpod_flutter: 3.5.0-beta.9 # in the Flutter package

Also bump the Dart SDK constraint in the root pubspec.yaml and <project>_server/pubspec.yaml to match the 3.5 minimum:

environment:
sdk: '^3.10.3'

From the project's root folder, refresh dependencies. Dart workspaces (used by projects created with the 3.3+ scaffold) resolve all sub-packages in one command:

$ dart pub upgrade

If your project doesn't use a Dart workspace (there's no workspace: block in the root pubspec.yaml), run dart pub upgrade separately in each sub-package. To adopt workspaces, see Dart's pub workspaces documentation.

Then refresh the generated server and client code:

$ serverpod generate

Generate the 3.5 migration

Version 3.5 adds a few new internal Serverpod tables and updates some indexes to greatly improve logs performance on Insights. Create a migration that captures these schema deltas so your database can be brought up to date:

$ serverpod create-migration --tag "upgrade-3.5"

This writes a new migration to <project>_server/migrations/. It will be applied to your database in the next step.

Adopt the new development workflow

Version 3.5 introduces a faster, integrated development workflow. The new serverpod start command runs your server, your Flutter app, and (optionally) your database in a single watch process with hot reload, replacing the manual docker compose up + dart bin/main.dart + flutter run triad. The result is a tighter edit-save-see-result loop and built-in tooling for migrations, hot restart, and agent skills.

Before running it, choose how to handle the database.

Choose your data store

You have two paths. Pick the one that fits where you are today; both work with serverpod start.

Keep your Docker Postgres (easiest upgrade)

If you've been developing against a Docker Postgres on 3.4, you can keep it without changing your config. Pass --docker to serverpod start so it uses your existing docker-compose.yaml:

$ serverpod start --docker

With --docker, serverpod start brings up Docker if it isn't running, and tears down the compose stack on exit if the command brought it up.

Version 3.5 ships a built-in Postgres that runs as a child process of your Dart server, using the same Postgres dialect as production. It has practical advantages over Docker for day-to-day development:

Pros:

  • No Docker dependency.
  • No TCP port conflicts (uses a Unix domain socket by default).
  • Cleanup by deleting the data directory.

Trade-offs:

  • Tests must run with --concurrency=1 (the cluster is single-tenant).
  • Manual access (e.g. with psql) requires a Dart process to be running the cluster.

To switch, add dataPath to the database section of <project>_server/config/development.yaml and <project>_server/config/testing.yaml:

database:
host: localhost
port: 8090
name: <project>
user: postgres
dataPath: .serverpod/dev/pgdata

For testing.yaml, use a separate directory, for example dataPath: .serverpod/test/pgdata.

Once dataPath is set, serverpod start uses the embedded Postgres automatically:

$ serverpod start

The dataPath setting belongs in development.yaml and testing.yaml only. For production, use a managed Postgres: Serverpod Cloud provisions one for you, or you can connect to a managed service like Cloud SQL or RDS. Do not add dataPath to production.yaml or staging.yaml.

warning

If you've added dataPath to your config and also pass --docker, the server connects to the embedded Postgres rather than your Docker Postgres. dataPath is honored by the server process regardless of the --docker flag, which only controls whether serverpod start brings up the docker-compose stack. If that's not what you wanted, your Docker volume is still intact: remove dataPath from your config to use the Docker Postgres again.

Start the server

The first run compiles the native build hooks (this can take about 30 seconds) and applies the migration you generated above. The server then starts and watches your project; saving a file hot-reloads the code.

Beyond the server, serverpod start also launches the project's Flutter app. By default it uses Flutter's web-server device and opens your browser. Pass --flutter-device <name> to target a specific device. For IDE debugging, projects scaffolded with 3.5 include a launch.json that runs serverpod start with the debugger attached; you can copy that file into your existing project from a fresh 3.5 scaffold if you want the same setup.

Add the agent skills (optional)

Version 3.5 ships AI agent skills (for editors like Claude Code and Cursor) and an MCP server. Install them with:

$ dart install skills

From your project's root folder, install the skills for your editor:

$ skills get --ide cursor

Replace cursor with the editor you use: antigravity, claude, cline, codex, copilot, cursor, opencode, or generic to install at the .agents folder.

Production deployment notes

Your production build needs to switch from dart compile exe to dart build cli. The 3.5 server includes native build hooks that dart compile doesn't support, and produces a bundle (executable plus its native libraries) rather than a single static binary, so your Dockerfile needs a few updates.

Copy the updated Dockerfile from the 3.5 framework template or a fresh 3.5 project's <project>_server/Dockerfile. The key changes vs. the 3.4 pattern: build from the project root (not the server directory), copy the bundle directory, update ENTRYPOINT to point at the bundled binary, and bump the Dart SDK base image to 3.10.x or newer.

What's new in 3.5

  • serverpod start TUI: hot reload on save, R to hot restart, M to create a migration, A to apply migrations, P to apply a repair migration.
  • Flutter app spawning from serverpod start so the Flutter app runs alongside the server in the same TUI.
  • AI agent skills and MCP server scaffolded during serverpod create; existing projects opt in with dart install skills and skills get.
  • Embedded Postgres: zero-Docker development via dataPath.
  • SQLite database support as an alternative dialect to Postgres.
  • Client-side database generation for the Flutter app.
  • jsonb column support with GIN index operator classes, and dynamic fields on models and endpoints.
  • unique keyword for simpler unique indexes in model files.
  • upsert and upsertRow on the ORM, and asc() / desc() convenience methods on orderable columns.
  • Recurring future calls via the new claim-based scheduling.
  • OAuth2 PKCE Flutter web redirect for sign-in flows.
  • Health endpoints on the built-in webserver.
  • IDE and agent selection in serverpod create.

Troubleshooting

Port conflicts on startup

Running more than one Serverpod server on the same machine can conflict on the default ports (8080 for the main server, 8090 for the database). This is a long-standing limitation, not specific to serverpod start. Stop the other server, or run on different ports.

Agent skills aren't picked up after install

Run skills get again from the project's root folder. Some editors, like Cursor, require enabling the MCP server in their settings.

Still stuck?

If something here didn't go as expected, reach out on the community page.

  • Migrations: how Serverpod's migration system works under the hood.
  • Build your first app: the hands-on tour of the 3.5 workflow if you want to see serverpod start from scratch.