Self-Hosted Open-Source Apps Deployed with ox: All 85

ox deployed 85 open-source apps on fresh Ubuntu servers from pinned commits, one server per app, on October 17, 2026. A separate test agent then used each app the way its kind of app is used: it signed in and did a main task of the app, used its API, or opened the pages of a static site. It also checked each app's services and backup after a redeploy, a backup and a reboot of the server.

No app ran in Docker. ox built each one on the server from its pinned commit and ran it with systemd and Caddy, plus the services it needs, such as Postgres, Redis or MySQL. PhotoPrism is the one build that did not compile from source: its config installs the project's own Linux package for that commit. A few other apps also download a font, a model or a prebuilt tool while they build. The table further down lists all 85 with the repository, the commit, the first deploy time and what the test agent did.

How did ox deploy and test each app?#

Each app got its own fresh DigitalOcean server running Ubuntu LTS, with the app alone on it. Every server had the size s-8vcpu-16gb, which is 8 vCPUs and 16 GB of memory. The size was the same for all 85 apps and was not picked for any one of them.

Each app was pinned to one exact git commit, so you can fetch the same code that ran. Then the run did these steps, in this order:

1. Create the project and save its variables, which starts the first deploy. ox builds the app on the server and starts it, and the run checks the app.
2. Back the app up.
3. Redeploy the same commit.
4. Reboot the server.

The app was checked again after each step. For 32 of the apps, the run also deployed a staging copy beside production.

Then a separate test agent, one that did not write the config, used the app the way its kind of app is used. It did the app's main job, not every feature. Here is what each kind got:

The agent also checked each app's services, and its backup for 80 of them. The four static sites and SearXNG had no backup check.

What did the 85 apps cover?#

Which stacks did they use?#

The stacks run from Django, Rails and Laravel to Go, Java and Rust. Each app is listed once, under the stack its config and source show.

Which services did they need?#

72 of the apps use at least one service that ox sets up beside them on the same server, and 13 use none. The count comes from each app's ox.toml.

LobeHub also has a service of its own, a Postgres with the pg_search extension. Dify has a second Postgres database for its plugin daemon. The 13 apps without a service include the four static sites.

How long did the deploys take?#

The first deploy time runs from creating the project until the app's own checks passed, so it includes the build. The median across the 85 apps was 357 seconds, just under 6 minutes. 33 apps took under 5 minutes and 66 under 10 minutes. 14 took between 10 and 20 minutes, and 5 took longer.

The redeploy of the same commit is timed the same way. The slowest deploys were big builds: SurrealDB compiles its Rust code on the server, and AFFiNE builds a native module and several web apps.

Which apps did ox deploy?#

Every app is below, from A to Z. Each row has the repository and the exact commit that ran, the first deploy time and what the test agent did. Each app's name links its guide, which shows the ox.toml that passed, the variables it needed and the run's screenshots.

Every app ran on a DigitalOcean server with 8 vCPUs and 16 GB of memory.

What was not tested?#

• A custom domain. Each app was served on the address ox shows for the project. The domains page covers adding one.
• A restore from a backup. The backups were taken and checked, and none was restored.
• E-mail delivery. Joplin Server, Mautic, Outline, Saleor and solidtime sent mail to a test inbox, and the test agent read it there. Delivery through a real mail provider was not tested.
• Answers from an AI model. The server had no AI model key, so LobeHub's chat reply was an error. The test agent checked that the chat was kept.
• The sign-up captcha of diaspora*. The run turned it off with SETTINGS_CAPTCHA_ENABLE=false, as the app's example config documents.

Where to go next#

How ox self-hosts your apps covers what ox puts on your server and which stacks it runs. To deploy a repo of your own, the quickstart goes from sign-in to a live app in five steps. The Plausible guide is one of the 85 guides, each with the config that passed, and the CLI reference lists every command. The backups page explains how ox backs up the databases it sets up.

Frequently asked questions#

Can I self-host these apps without Docker?#

Yes. In these runs ox built each app on the server from its pinned commit (PhotoPrism from its own Linux package) and ran it with systemd and Caddy. It installed the databases the app needed beside it. Some of these projects publish Docker images or Compose files. ox ran none of them and installed no Docker.

Which open-source apps can ox deploy?#

The table above lists the 85 apps that ox deployed and the test agent used, including Mastodon, Odoo, Grafana, WordPress, Keycloak, Jellyfin and Plausible Analytics. Each row links the repository and the commit that ran. An app that is not in the table was not part of these runs.

What server did the apps run on?#

Each app ran alone on a fresh DigitalOcean server of size s-8vcpu-16gb (8 vCPUs, 16 GB of memory). After the reboot, the whole server used a median of 1,023 MB across the 84 apps measured (620 to 8,565 MB). ox works on any server from 1 GB up and tells you when memory is why an app stopped.

How long does it take to deploy an open-source app with ox?#

The median first deploy across the 85 apps took 357 seconds, just under 6 minutes, and 66 of the 85 finished in under 10 minutes. The slowest, SurrealDB, took about 69 minutes because it compiles Rust on the server. A redeploy of the same commit had a median of 170 seconds.

New to running apps on your own server? How ox self-hosts your apps shows what ox sets up on your server.