> ## Documentation Index > Fetch the complete documentation index at: https://docs.levelrail.com/llms.txt > Use this file to discover all available pages before exploring further. # Server profiles > Choose minimal, standard or suite to decide which optional parts of the control plane start, and switch later without losing data. A profile is one setting that decides which optional subsystems start. The default, `standard`, is the install you have always had. `minimal` keeps a one-server install as light as Dokku. `suite` turns on everything for power users. Idle CPU and memory were never the problem (see [Performance](/performance)). What a profile changes is how many things run, listen, warn and grow your disk, so there is less to explain on a small box. ## What each profile starts | | `minimal` | `standard` (default) | `suite` | | - | - | - | - | | Mesh and WireGuard | off | off | on | | Plugins and marketplace | off | off | on | | Experimental features | off | off | all on | | Metrics and log retention | 3 days | 15 days (7 days on a disk under 20 GiB) | same as standard | | Backup scheduler and WAL shipping | start when a backup is scheduled | start at boot | start at boot | | Agent listener (port 9443) | loopback only until a remote node exists | all interfaces | all interfaces | | Security score | backup and key rotation are recommendations, not score items | scored | scored | | Doctor RAM and CPU floors | 512 MiB and 1 core | 896 MiB and 2 cores | 896 MiB and 2 cores | | Build SBOMs | off | off | on | | Host disk, memory and patch history | with the mesh | with the mesh | on (mesh is on) | A pipeline engine that only polls while a run is active, a retention default that follows disk size, and the corrected doctor floors apply to every profile. ## Choose a profile At install time: ```bash theme={"dark"} curl -fsSL https://levelrail.com/install.sh | sudo sh -s -- --profile minimal ``` `LEVELRAIL_PROFILE=minimal` does the same. The installer writes `Environment=APP_SERVER_PROFILE=minimal` to the systemd drop-in and keeps it on upgrades. A machine with under 2 GB of RAM is told that minimal fits it, but is never switched silently. Later, from **Settings, Beta features, Server profile**, or from the CLI: ```bash theme={"dark"} levelrail-cli settings profile get levelrail-cli settings profile set suite --restart levelrail-cli settings profile set minimal --dry-run ``` Only an administrator can change it. The choice is saved at once and applies at the next restart of the control plane, because most subsystems are wired at startup. `--restart` (or the **Restart now** button) asks for one. ## Which setting wins 1. `APP_SERVER_PROFILE` in the server environment. When set, the dashboard and CLI show it as pinned and refuse to change it. 2. The profile saved from Settings or the CLI. 3. `standard`. Inside a profile the same order holds for each part: `APP_MESH_ENABLED` beats the mesh toggle in Network settings, which beats the profile. `APP_EXPERIMENTAL` beats a switch saved on the Beta features page, which beats the profile. A retention variable (`APP_METRICS_RETENTION`, `APP_LOGS_RETENTION`) beats any profile default. ## Moving between profiles never deletes data Moving up only starts things. Moving down stops loops and makes gated routes return 404, and keeps every row. A few moves are checked first: | Move | What happens | | - | - | | Down to `minimal` with remote nodes enrolled | Refused. Minimal binds the agent listener to loopback, so remote agents could not reach it. Remove the nodes or choose standard. | | Away from `suite` while the profile is the only reason the mesh runs and nodes use it | Refused. Pin the mesh on in Network settings or remove the nodes. | | Down while plugins are enabled | Refused. Disable them first so their containers are not left unsupervised. | | Down to a shorter retention | Needs a confirm (`--confirm`), because the next hourly sweep deletes samples and log lines older than the new window. | `--dry-run` prints the same checks without saving. The profile is stored in the control plane database, so restoring an older snapshot restores the older profile. ## Tuning knobs | Variable | Default | Meaning | | - | - | - | | `APP_SERVER_PROFILE` | unset | Pins the profile to `minimal`, `standard` or `suite`. | | `APP_PROFILE_MINIMAL_RETENTION` | `72h` | Retention on the minimal profile. | | `APP_SMALL_DISK_BYTES` | 20 GiB | Data disks below this get the shorter default retention. | | `APP_SMALL_DISK_RETENTION` | `168h` | That shorter retention. | | `APP_LOW_RAM_BYTES` | 2 GiB | Memory below which the dashboard recommends minimal. | | `APP_PROFILE_LAZY_START_INTERVAL` | `5m` | How often minimal checks whether a deferred loop has work. | | `APP_PIPELINE_IDLE_INTERVAL` | `1m` | Pipeline engine poll while no run is active. | `APP_PROFILE` is a different setting: it names the CLI credentials profile. ## What it costs Measured with the real boot harness on one machine (the table is in [Performance](/performance#server-profiles)): `minimal` runs 116 goroutines and 35 loops, `standard` 118 and 38, `suite` 128 and 41. The difference is small by design: the process was already light at about 162 MB. The gain is fewer things running, listening, warning and filling the disk, not a smaller process. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.