Flaem

Security for accounts, staff and nodes in a hosting panel

Flaem protects client accounts with two-factor sign-in and passkeys, your team with per-action permissions and a two-person rule, your money with fraud screening and your network with abuse detection. Traffic to the nodes is encrypted and every release is signed.

2 people

approve deleting a service or a node and moving money

16 chars

in every backup code, each valid once

30 min

is the longest a view-as-client session lasts

5 tries

a day for the account recovery code

01Sign-in

Sign-in with a second factor, passkeys and a way back in

Clients sign in with a password and a second factor, with a passkey and no password, or with a Google, GitHub, Discord, Steam or any OpenID Connect account. Each method has its own role and its own way back when the phone is lost.

Instead of a password

  • Passkey
  • Google, GitHub, Discord, Steam, OpenID Connect

Second factor

  • Authenticator app
  • Passkey

Emergency route

  • Backup codes
  • Account recovery code
You switch each sign-in provider on separately
Client account security: two-step verification, passkeys, recovery codes, active sessions and sign-in history
A cloned passkey
A passkey counts its signatures. When the counter does not move, two copies of the key are in use: the sign-in is refused, the passkey is disabled on the spot and its owner is told.
Backup codes
They are created together with two-factor sign-in, shown once and stored only as hashes. Each works once, and a new batch replaces the whole old one, so a printout never mixes valid and used codes.
Steam accounts
They have to confirm an email address before buying anything.
02Recovery code

The recovery code treats every use as an account takeover

The recovery code works when the password and every second factor are gone. It is shown once, at the first sign-in, and belongs on paper or in a password manager. When someone uses it, Flaem assumes the worst and clears the account of everything an attacker could have had.

Someone enters the recovery code
  1. 1A new password has to be set
  2. 2Every session and API key stops working
  3. 3Open reset links and address changes are cancelled
  4. 4The authenticator app and backup codes are removed
  5. 5Passkeys are removed
  6. 6The recovery code is replaced with a new one
  7. 7The owner is told on every channel they have set up

Five wrong codes for one username or address within a day and further attempts are refused. The limit counts what was typed, not the account, so a refusal does not reveal whether such an account exists.

03Sessions and alerts

Sessions under control and a notice for every important change

Clients see the list of devices they are signed in on and sign out everywhere with one button. They get a message straight away about changes that could be the start of an account takeover.

Security notification

Someone signed in to your account from a new device.

Sign out everywhere
The session list shows the devices, and one button ends all of them at once.
Password change
It also breaks the account's connection to Discord.
04Team and permissions

Permissions for every action and mandatory two-factor sign-in for staff

Every screen and every action in the panel has its own permission, so support can read invoices without being able to issue refunds. You build the roles yourself, down to a single action.

ActionSupportBillingInfrastructure
Reading a client invoice
Issuing a refund
Viewing an account as the client
Restarting a node
Deleting a node
Changing plan prices
CanCannotWith a second signature

An example split of roles

A new route cannot be a hole
The permission a route needs is written in the same place the route is added to the panel. Tests check that nothing under the admin panel is left open, and a route without an entry is refused.
Mandatory second factor
Once you switch it on, a team member without a second factor can still sign in but only sees the screen for adding an authenticator app or a passkey. Nobody gets locked out of their own panel by it.
What counts as a second factor
An authenticator app or an active passkey. Backup codes and the recovery code are emergency routes, so they do not meet the requirement.
Staff role editor: permissions for single actions, a preview of what the role will see and the actions that need a second signature
05Two-person rule

A two-person rule for actions that cannot be undone

Deleting a service, deleting a node and moving money turn into a request that another qualified person has to approve. The approval carries out the stored action straight away, so nobody has to come back and click again.

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

First person

Requests the action. The panel records exactly what it targets.

Request

Waits in the queue and on the on-call screen, phone included.

Second person

Approves or rejects. Approving carries out the stored action.

You decide what needs a signature
The list of actions that need a second signature is a setting, not code. You can add to it without waiting for a release.
An approval cannot be redirected
The request keeps the target it was created for, so approving it cannot hit a different service or node.
A one-person team
When there is nobody else who could approve, the action goes through and the audit log records that one person did it. The safeguard never locks out the only operator.
Clients are not part of it
A client cancelling their own server needs no second signature. The rule is about staff acting on other people's things.
06Audit log

Audit log: who, what, when, from where and on whom

Every staff action leaves an entry with the person, the action, the time, the address it came from and the target: an account, a service or a machine.

Who

a team member or a reseller

What

the action and its details

When

the exact time

From where

the address the request came from

On whom

an account, a service or a machine

An address that cannot be faked
The client address header is only trusted from a reverse proxy you list as trusted. A direct request cannot write someone else's address into the log.
Viewing an account as the client
The start, the end and every request during the preview get their own entry with the person viewing and the account viewed.
07Fraud screening

Fraud screening: an order accepted, held for review or rejected

Game hosting attracts stolen cards: the service is ready in minutes and the chargeback arrives weeks later. Flaem screens every order and every sign-up before anything is provisioned.

Order or sign-upScreening
Your scoring provider
On top of your own rules you can add a score from MaxMind minFraud or FraudLabs Pro.
No double resources
An approved order gets exactly the resources the sale reserved, at the price from the moment of purchase. Nothing is bought twice.
What the rules do not punish
A new client or the size of the order. Every honest client has a first order, and in hosting a fraudster tends to buy many cheap servers rather than one expensive one.
08Abuse

Abuse detection: miners, spam, scanning and outgoing attacks

An operator answers for every packet that leaves their address space. Flaem notices what client machines do to others, takes in abuse reports and reacts step by step, and the client always knows what was noticed.

What Flaem notices

Miners
sustained CPU use
Spam
connections to mail servers
Scanning
knocking on many ports and addresses
Outgoing attacks
packets per second, UDP share, number of targets and SSH and RDP password guessing
Abuse reports
complaints taken in from outside and from staff

How it reacts

  1. Observe1/4

    The finding goes on the case. Nobody is told.

  2. Alert2/4

    The team gets an alert and the client a message about what was seen and that nothing has been changed.

  3. Throttle3/4

    Outgoing mail only: the machine's firewall blocks the mail ports ahead of the client's own rules.

  4. Suspend4/4

    The service is suspended with the reason "abuse", and reinstating it restores its state.

The reaction only climbs
While a case is open, the reaction can only go up. A miner that goes quiet for an hour stays suspended until a person has looked. A case that only ever alerted closes by itself after a quiet day.
A verdict with numbers
Every verdict carries the measurements it was based on, so the client can appeal and you can tune the threshold.
Every reaction can be undone
Every action is explained to the client, recorded for the team and in the audit log, and reversible by staff.
Captcha and rate limits
Cloudflare Turnstile or hCaptcha at sign-up and in the places you choose. Every public API address has its own limit, and a refusal says after how many seconds to come back.
09Nodes and packages

Encrypted traffic to the nodes and signed releases

The most valuable secrets of an installation travel between the control plane and the nodes: root passwords of new VPS machines, SFTP logins and backup addresses. That is why this channel is always encrypted and the node knows who it is talking to.

  1. 1

    Release key

    The publisher signs every package and release manifest with an Ed25519 key.

  2. 2

    The panel checks the signature

    The panel only trusts a manifest whose signature matches the key built into the program. The download address alone proves nothing.

  3. 3

    A person starts the update

    The panel checks for new releases itself and shows the steps, but a person starts the update.

  4. 4

    The node rolls back if something fails

    A node updates itself, and when the new version does not come up, it goes back to the previous one.

Certificates without the hassle
By default the control plane issues its own certificate and the node pins its certificate authority during installation. You can also supply your own certificate for an address with a real name.
Separate credentials
Every node has its own credentials, so taking over one does not open the rest of the fleet.
10Vulnerabilities and incidents

What happens when someone finds a vulnerability or an incident happens

Reports go to hello@flaem.net. Once a problem is confirmed, operators of the affected installations hear from us within 24 hours, and the fix arrives as a normal signed release.

  1. Report

    Email a vulnerability or a suspected incident to hello@flaem.net. A description, the panel version and the steps that lead to the problem are enough.

  2. Confirmation and assessment

    We reproduce the problem and work out which versions and which parts of the panel or the node it affects, and whether data could have left the installation.

  3. Notice to operators

    24 h

    Within 24 hours of confirmation, operators of the affected installations get an email at the address tied to their licence.

  4. Signed release with the fix

    Ed25519

    The fix ships through the Release channel and, like every release, is signed with the Ed25519 key. The panel checks the signature and shows the release notes on the update screen.

  5. Changelog entry

    Once the fix is out, the public changelog records what was affected and which version fixes it.

What the notice to operators contains

1
What is affected
The versions, the parts of the panel or the node, and the conditions in which the problem shows up.
2
What to do now
A workaround or a setting that closes the problem until you update, and the version number that carries the fix.
3
Whether data could be exposed
Which data and whose could have been exposed, or a plain statement that none was.
The fix takes the same road as every release
There is no separate channel for urgent patches. A fix carries the same signature and passes the same checks, so nobody can slip you a fake "urgent update".
An overdue licence does not block the fix
The node update command keeps working even after an overdue licence has put the panel into read-only mode, because it may carry a security fix.
Installations on an offline token
They never contact us, so news of a problem reaches them by email only.
11Client data

Data export and account closure under GDPR

Clients download their data as JSON or CSV and can close their account. Closing it complies with GDPR and does not destroy the accounting records you have to keep.

Export
  • JSON
  • CSV
Account closure
RemovedThe account and data the books do not need
KeptInvoices and accounting records
12FAQ

Questions about Flaem security

Sign-in, staff, abuse, nodes and reporting vulnerabilities.

Yes, with a passkey or with a Google, GitHub, Discord, Steam or any OpenID Connect account. You switch each provider on separately.

They sign in with a backup code. If the password is gone too, they use the account recovery code, which clears the account of every session, key and second factor.

Yes. A team member without a second factor can sign in but only sees the screen for adding one.

No. When there is nobody else who could approve, the action goes through and the audit log records it.

Only according to the policy you set: from observing, through alerts and mail throttling, to suspension. Staff can undo every reaction, and the client knows what was noticed.

The panel does it for you: a release manifest has to carry a signature that matches the key built into the program, or the panel will not accept it.

Email hello@flaem.net with a description, the panel version and the steps that lead to the problem. Once it is confirmed, operators of the affected installations hear from us within 24 hours, and the fix ships as a signed release.

See Flaem security in a demo

Tell us how many people are on your team and what you sell. We will show permissions, the two-person rule, fraud screening and abuse detection, then work out your licence from the price list.