Flaem

Hosting infrastructure run from the panel, not over SSH

You install the panel with one command and every node with a command the panel writes for you. After that, updates, network changes, installation backups, service moves and node failures are handled from one place, and every risky change comes with a way back.

Shared address for the panel and API
Control plane 1
serving
Control plane 2
on standby
Encrypted node channelcommandstelemetry
Proxmox node

VPS, LXC, games and web

Proxmox node

VPS, LXC, games and web

Container node

Games and web

agent update
Container node

Games and web

01/04

A node agent updates from the panel and returns to the previous version if the new one does not come up.

The control plane is the panel, the API and the database. Nodes are the machines your customers' services run on.

01Installation

The panel installs with one command

The installer puts the panel, the API, the database and the cache on your machine as containers. It asks a few questions and does the rest itself, including the HTTPS certificate and the first administrator account.

curl -fsSL https://flaem.net/releases/stable/install.sh | sudo bash
Releases from flaem.net/releases, stable channel.

The installer asks for

  1. 01

    Directory, name and address

    Where the installation lives on disk, what it is called and which domain your customers will open the panel on.

  2. 02

    HTTPS and ports

    With a domain, the Let's Encrypt certificate is issued on its own. You can also terminate HTTPS on your own proxy.

  3. 03

    Licence key

    If you already have one. Without it the installation starts a 14-day trial with every feature on.

  4. 04

    The first administrator

    Username, email and password of the account you will sign in with first.

It does the rest

  • Checks the machine before writing anything
  • Installs Docker if it is missing
  • Generates every secret the installation needs
  • Runs database migrations as a separate, visible step
  • Creates the first administrator account
  • Starts the panel on your domain with HTTPS
The panel is live on your domain

The database is not exposed

PostgreSQL and Redis are not published outside the machine. Nodes reach the control plane on a single port, 50051.

Sign-ups wait for you

Until you sign in for the first time, nobody can create an account in your store.

02First run

A 14-step wizard takes you through the first run

After your first sign-in the panel lists everything the installation needs before customers can buy: from the licence and invoice details to the first published offer. Eight of these steps are enough for the first sale.

0 of 14 donecannot be skippednot needed

Account and company

  1. 01

    Licence

    A key or a signed token. Until then the installation runs on its trial.

  2. 02

    Second factor for staff

    Two-factor authentication required for every team member.

  3. 03

    Seller details

    cannot be skipped

    Company name, tax number and address on invoices. Nothing sells without them, so this step cannot be skipped.

Money and email

  1. 04

    Public address and outgoing email

    The panel domain and the SMTP server your customer emails go out through.

  2. 05

    Currency and tax

    Base currency and VAT rates. The currency can change until the first sale.

  3. 06

    Payment gateways

    Each of the 7 gateways is switched on by pasting its keys.

  4. 07

    KSeF

    KSeF token and environment, if you invoice in Poland.

Channels

  1. 08

    SMS messages

    not needed

    An SMS provider, or marking the step as not needed, which you can undo at any time.

  2. 09

    Domain registrar

    Your registrar account. The panel checks the connection by asking for the balance, without registering anything.

  3. 10

    Push notifications in apps

    Push to customer phones. Without this step, notifications reach the browser.

Selling

  1. 11

    IP address pools

    Addresses the panel assigns to new services.

  2. 12

    Nodes

    The first machine, added with a command the panel prepares.

  3. 13

    Packages for sale

    Resources and prices of what you sell.

  4. 14

    Store

    A published offer customers can see and buy.

Flaem setup checklist: the remaining steps and the domain registrar settings

Read from configuration, not ticked by hand

The wizard never asks you to tick anything. The panel reads the state of every step from what is actually configured, so the list cannot drift from reality.

A wizard for everything new

The same rule holds after launch. Every action that adds something, such as a node, a service or a package, opens a step-by-step wizard.

03Nodes

You add a node with a command the panel prepares

In the panel you enter the machine name and address, and the panel generates the command to paste. Before anything is installed, the machine is checked from both sides at once: by the node from the inside and by the panel from the outside. You see both results in one place.

The node checks itself

  • System: Ubuntu 22.04+, Debian 12+, Rocky 9+ or AlmaLinux 9+
  • Kernel 5.10+ and cgroup v2 with CPU, memory, disk and process limits
  • Docker 24+ with the overlay2 driver
  • Memory from 2 GB, 4 GB recommended
  • At least 10 GB of free disk, 20 GB recommended
  • Bare metal or KVM, no OpenVZ or LXC
  • A clock that matches the control plane
  • A connection to the control plane channel

Result in the panel

Every check on its own, before a single file is installed. You can run the check alone even on a live node, because it changes nothing.

The panel checks it from outside

  • Whether the machine answers at the given address
  • Whether service ports are reachable from the internet

A node cannot see whether its ports are reachable from the internet, and the panel cannot see its cgroups. That is why both sides check it.

  1. 01

    Check

    Every test runs and reports to the panel before anything is installed.

  2. 02

    Install

    Dependencies first, then each node agent separately, each with its own result. You know at once which part did not come up.

  3. 03

    Join

    The node joins the control plane only once everything runs. If something failed, you fix it and run the same command again.

The node mode picks itself

On a Proxmox VE host the installer picks Proxmox mode by itself. On any other machine the node runs game servers and websites.

Node installer, automatic mode

Proxmox mode

On a Proxmox VE 8 or 9 host.

  • KVM VPS
  • LXC containers
  • Game servers
  • Websites

Container mode

On a plain Linux server.

  • Game servers
  • Websites
A separate credential for every node
Each node reaches the control plane with its own secret and certificate. You can rotate, disable or revoke them, and the customer servers on that machine keep running.
Signed packages
Nothing a node downloads is unpacked until its Ed25519 signature checks out. A truncated download ends as a signature error, not as an agent that refuses to start.
Single-use token
The command carries a token valid for 24 hours that works once. A new command for the same node cancels the old one, and the script never stores the token.
04Updates

Panel and node updates from the panel, with automatic rollback

The panel checks for new releases itself and shows the steps for your exact version jump. You start the update, in a window you choose. If the new version does not come up on a node, the node goes back to the previous one without you.

New releaseMaintenance windowInstallVerify
New release

Panel and control plane

The panel compares its version with the releases and shows the steps for your jump, including whether a database migration is needed. The check reads a static file next to the releases, not an API, so nobody can see who runs which version. On an isolated network you turn it off with one setting.

Agents on nodes

Each agent is stopped, swapped, started and verified in turn. If it does not come back, the node restores the version it was on. Every package passes a signature check before it is unpacked.

Node operating system

Every node shows pending Proxmox and Debian packages, security fixes separately, an old running kernel and whether a reboot is needed. You plan a window, the node's customers get a maintenance notice, and you can move machines away first. Afterwards you get a report of what changed and whether everything came back.

05Templates and images

Fresh OS templates, known image versions and machines on standby

A new VPS starts from an OS template and a game server from a container image. Flaem keeps templates current and tested, tells you which image version every server runs, and keeps the most popular packages ready before anyone orders them.

Service panel without a terminal: apps with a domain and HTTPS, backups and the install wizard

Template refresh with a test boot

Each template shows which image release it was built from and whether a newer one exists. One click does the rest, and the old template stays for rollback.

  1. 01

    Download

    The new OS image release.

  2. 02

    Checksum and signature

    The file must match its checksum and the publisher's signature.

  3. 03

    Build

    The new template is built next to the current one.

  4. 04

    Test machine

    It has to pass cloud-init and report its agent.

  5. 05

    Swap

    Only now does the new template replace the old one.

current versionolder version

Image versions on nodes

You see which servers on which node run which image version, with a summary for every node.

Repositories on the internet
Mirror in your location
Customer VPSes

Local package mirror

Copies of the OS repositories sit in your location, so installs and updates on customer VPSes stay local, run faster and use no outside transfer. You turn the mirror on per location.

readybeing preparednew order

Ready machines on standby

A node keeps a few VPSes and containers prepared in advance for the most popular packages, so a new order starts in seconds instead of minutes. You set how many spares to keep, so they do not hold space for nothing.

06Node network

Change node networking from the panel while a kernel filter drops junk traffic

A network change is the most common way an operator locks themselves out of a remote machine. In Flaem every node network change has an automatic way back, and a filter in the node kernel stops junk packets before they reach customer servers.

Network changes with automatic rollback

You add a bridge, a VLAN, bonding, MTU or an address and see the diff before saving. After saving, the node has a set time to confirm it can still reach the control plane. If it does not, it returns to the previous configuration on its own. The same change can go to many nodes at once.

Diff before saving
  • +New bridge for customer networks
  • +VLAN on the uplink to the switch
  • ~Larger MTU on the network card
The node confirms it can reach the control plane
Waiting for confirmation

Traffic filter in the node kernel

The filter runs in the node's kernel. It drops junk traffic before it reaches the servers, answers server list queries from memory and limits packets per source address. You switch it on and off per node, so you decide where it runs.

Kernel filterCustomer servers
  • Junk packetsdropped before the server
  • Server list queriesanswered from filter memory
  • Player trafficreaches the servers
07Node memory

Node memory under control instead of oversold blind

You sell more memory than a node physically has, because not every VPS uses all of it at once. Flaem shows how much you really save and how close the node is to running out, and when the kernel has to stop a machine, it records the reason instead of leaving a mystery.

Memory sold in packages
Physical node memory
Used by machinesReclaimed by page mergingHeadroom before running out
One VPS with ballooning
guaranteed minimumsold maximum
VPS stoppedReason: node out of memoryOne alert to the operator
  1. 01

    Kernel same-page merging (KSM)

    You turn it on per node, and the node merges identical memory pages from many machines into one. The panel shows how many gigabytes this really saves.

  2. 02

    Ballooning in every package

    A package has a guaranteed minimum and a sold maximum. A machine gives back memory it does not use and gets it back when it needs it.

  3. 03

    A stop with its reason

    When the node kernel stops a VPS for lack of memory, the panel records it as the stop reason and you get a single alert. Nobody has to guess why the machine shut down.

08Node failure

When a node goes down, a person decides and Flaem does the rest

Nothing moves on its own in the middle of the night. The panel waits as long as you set, alerts whoever is on call, and then spreads services across the other nodes with one command so that none goes over your threshold.

threshold, for example 90%
not responding
not responding
not responding
not responding
not responding

Rebuilding a node on new hardware

After a hardware failure you point at a new machine, and Flaem installs everything on it and restores services from backups with the same addresses and ports.

Hardware failureNew machine, same addresses and ports

Moves with seconds of downtime

A VPS disk copies in the background while the machine runs, and changed blocks follow in further passes. At the end there is a short stop and a start on the other side. Game servers and websites move the same way: only the difference is copied in the window, and players get a notice. An error before the switch leaves the service where it was.

Whole disk
Changed blocks
Later changes
Switch-over
machine runningseconds of downtime

The panel shows the phase and the expected downtime, the customer is informed, and the server name moves with it.

On-call view in the phone app: alarms, the queue and node health
09Control plane

Two control planes, installation backups and maintenance without service downtime

The control plane is the panel, the API and the database. On a single machine its failure stops sales and the panel. Flaem runs two control planes behind a shared address, backs up the whole installation on a schedule, and shows customers a notice during maintenance while their servers keep running.

  1. An ordinary day

    01
    Load balancer or shared address
    Control plane 1
    serving
    Control plane 2
    on standby
    Nodes and customer serversrunning

    One control plane serves traffic, the other keeps a live copy of the database and waits.

  2. Control plane failure

    02
    Load balancer or shared address
    Control plane 1
    not responding
    Control plane 2
    serving
    Nodes and customer serversrunning

    Traffic moves to the second control plane. The store, payments and the API keep working.

  3. Maintenance

    03
    Load balancer or shared address
    Control plane 1
    maintenance
    Control plane 2
    maintenance
    Maintenance in progress
    Nodes and customer serversrunning

    The panel shows customers a maintenance notice while their servers run without interruption.

Backup of the whole installation

Accounts, invoices, the licence and sealed credentials, in other words everything that is not in customer server backups.

  • Daily at a time you pick, to a separate disk or S3 storage, apart from customer backups
  • Retention by count, days and size, and the last copy is never deleted
  • Verify a copy with one button: a read-back that writes nothing
  • Passwords and integration keys stay sealed in the copy, so the copy alone does not leak them
  • A restore asks you to type the installation name, and the attempt is logged before it starts
  • A warning when the newest good copy is too old
10Installation health

Installation health, early warnings and agreement with nodes

The panel checks every part of the installation every minute and answers on one screen what works and what does not. It separately warns about what is coming, and compares its records with what actually runs on the nodes.

Flaem operator dashboard: sales and MRR, live nodes and the issues waiting for a person

Five states that never blur into one

  • HealthyAnswers and works correctly.
  • DegradedAnswers, but something is off, for example a node whose telemetry cannot be reached.
  • DownProbed and silent. The alarm opens after two such results in a row.
  • UnknownNobody has checked yet, or nobody could.
  • Not configuredNot a failure. An installation without an SMS gateway simply does not have one, so no alarm opens.

Alarms go by email, Discord, Telegram and SMS, each channel with its own minimum severity.

Early warnings

WarningWhen it opens (default setting)
No recent installation backupThe newest good copy is older than 48 hours, the last one failed or the schedule is off
Node running out of disk or memoryLess than 10% free disk, less than 5% free memory, or a fill rate that reaches full within 7 days
Licence expiring14 days and 3 days before the paid period ends
Licence server not answeringAfter 3 days without an answer
Outdated node agentThe agent is older than the current release requires
Node clock skewA difference of more than 5 minutes

You change thresholds in settings, and zero turns a warning off. Each one stays a single warning while the problem lasts and closes on its own.

Panel recordOn the node
no owner

State reconciliation

The panel compares its records with what really runs on the nodes and shows the differences, including services without an owner.

Load on every node core

Node telemetry

Every core, disk and network card of a node, the connection table and storage fill, down to one-second resolution.

11Team and languages

A team with permissions for every action and a panel in five languages

Every team member sees and changes only what you allow. The most dangerous actions need a second person, and every change lands in the audit log.

Example role
Customer support
AreaSeesChanges
Customer accounts
Customer wallet
Customer services
Nodes
Invoices and refunds
Tickets

A permission covers every item a screen shows and every action, separately for viewing and for changing.

Two-person rule

Irreversible actions run only after a second person approves them.

  • Deleting a servicerequestsapproves
  • Deleting a noderequestsapproves
  • Moving moneyrequestsapproves

Five languages and your own translations

The panel, emails, the help centre and failure hints work in Polish, English, German, Czech and Ukrainian. The translation editor lets you change any text, and you can add your own language next to the built-in ones. Missing texts then show in English.

PLENDECSUKYour language
Audit log
Every change with the account that made it, the time and the object it touched.
Mandatory second factor
You require two-factor authentication from the whole team.
Roles scoped to nodes
A role can cover only selected nodes. Its holder will not see the others.
12Questions

Questions about operator infrastructure

Machines, updates, failures and backups: what operators ask before moving to Flaem.

No. The panel and nodes are separate machines, because both installers need the same ports. A modest machine is enough for the panel and its database: from 1 GB of memory, 2 GB recommended.

Their servers keep running. Disabling and revoking credentials only affect the node's connection to the control plane. If you want to stop selling on a node, you stop new services on it and the existing ones stay.

No. You start agent and system updates yourself, in a window you choose. Only the return to the previous version, when a new one does not come up, is automatic.

Not without you. The panel waits the time you set and alerts whoever is on call, and a person starts the move, for example with the "spread sensibly" command.

To a disk you choose or to S3-compatible storage, separately from customer server backups. The disk has to be a different one from the database's, so a backup cannot fill up the installation.

Yes. The node installer has a check-only mode that changes nothing and works even on a node already in service.

Releases are built for x86_64 processors, for the panel as well as for nodes.

See how Flaem runs your nodes

Tell us how many machines you have and what manages them today. In the demo we go through installation, adding a node and an update on your own case.