Running your own
Administration
The operator's side of VulnRadar: who can reach the admin panel, what each tab is for, how a runtime setting resolves, and how the operations you would otherwise reach for a shell to do (backups, blocklists, retention, mass email) work from inside the app.
- Staff roles
- 8
- Admin destinations
- 21
- Written to the audit log
- Every change
Who gets in
The very first account created on a fresh instance becomes the super admin. That role cannot be granted through the panel afterwards, which is what stops an escalation path existing at all: there is exactly one way to become super admin, and it is being first.
Everyone else is invited. An admin sends a staff invite for a specific email and a specific role; the link expires in seven days, the token is stored hashed rather than in plain text, and a pending invite can be revoked by deleting it. Sending an invite requires re-entering your password, as do the other 20 destructive admin actions.
Staff roles
Roles are not a single ladder. Four of them sit at the same rank as each other and differ only in what they are for, so you can give someone billing access without giving them session and 2FA controls.
| Role | Scope |
|---|---|
| support | Read-only. The user directory, scans, reports and subscriptions, plus the support ticket inbox. No audit log and no mutations. |
| billing | Plans, subscriptions and the billing overview. Cannot see sessions, the audit log or error logs. |
| security_analyst | Sessions, 2FA resets, moderation, reports and the audit log. No billing, settings or backups. |
| content_manager | Moderation, reports, announcements and notifications. No sessions and no audit log. |
| ops | System stats, error logs, scans, engine feedback and cache control. Deliberately has no access to the user directory. |
| moderator | Everything the specialists cover for user management: edit, disable, sessions, 2FA reset, delete any scan, audit log. Not impersonation, not settings, not backups, not deleting users. |
| admin | Everything except the super-admin-only reserve. Settings, backups, broadcasts, invites, impersonation. |
| super_admin | Everything, plus the sole ability to apply an in-place update. Cannot be modified by anyone else. |
What each tab does
The panel has 21 destinations in 7 groups, and each one is filtered by your role, so what you see depends on what you hold. Groups are the question an operator is trying to answer, not the kind of row the data is, and they are ordered by how often that question gets asked. The Overview tab is the one to check first: it is a worst-first health list covering the scanner queue, failed scans in the last day, when the last backup ran, errors logged in the last hour, failed emails, unresolved security alerts, tickets waiting on staff, and pending staff invites.
| Group | Tabs |
|---|---|
| Operations | Overview, Scanner Queue, Error Logs, Email Logs, Backups, Updater |
| People | Users, Teams, Staff, Audit Log |
| Support | Tickets, AI Chats |
| Trust & Safety | Security Alerts, Access Rules, Blocked Data, Hosts & Shares |
| Messaging | Broadcast, Notifications |
| Billing | Revenue |
| Configuration | Settings, Engine Feedback |
Two are worth knowing about before you need them. Error Logs captures every console.error the application emits, with credentials, API keys, tokens and email addresses scrubbed before storage, so you can diagnose a production failure without shell access. Email Logs records every outbound send attempt with a redacted preview that never contains a working link or code; its status is the mail server’s accept or reject, which is not the same as delivered.
The settings registry
Almost everything configurable is a typed entry in one registry rather than an environment variable you have to redeploy to change. Each entry declares its type, its bounds, the group it belongs to, and help text, and its default is the same constant the code ships with, so the form cannot drift from what the application actually does.
How a value resolves
Database value, then environment variable, then the shipped default. The database wins because that is the layer the admin panel edits. Values are cached for 30 seconds, and saving invalidates the cache for the process handling the save, so you see your own change immediately while other processes converge within the window.
If the database is unreachable, resolution fails open: every key falls back to its environment or shipped value rather than everything reading as off. A database outage should not also flip every feature flag.
Runtime settings and build settings
Most settings take effect within the cache window. A minority are marked as build-tier, meaning the running application reads the value compiled into the build: saving one records your intent but does not change behaviour until the next build. The panel labels these, and it is the single most common source of “I changed it and nothing happened”.
Resetting a setting deletes its row rather than writing the current default back into it. That matters on upgrade: a reset key keeps following the shipped default, while a key you explicitly set stays pinned to your value across releases.
Audit log
Every staff action writes a row: who did it, to whom, what, from which address, and when. That includes settings changes and resets, user actions, broadcasts, blocked-data operations, backups, staff invites and revocations, impersonation start and stop, and exporting the audit log itself.
Email addresses in the free-text detail are masked before the row is written, so a log kept for a year does not become a store of addresses belonging to deleted accounts. Reading the log needs the audit permission, which the security analyst role has; exporting the whole table as CSV or JSON requires full admin, and the export is itself logged.
Impersonation
An admin or super admin can sign in as a user to reproduce a problem. It is deliberately narrow: only plain user accounts can be targeted, never staff; it requires re-entering your password; the impersonated session is capped at one hour regardless of your normal session length; and a banner is visible across the whole application for the entire time.
Your own session is held in a separate cookie and restored when you stop. Both the start and the stop are written to the audit log, and the stop is attributed to you rather than to the account you were inside. If your original session expired while you were impersonating, stopping ends the impersonated session anyway and asks you to log in again, so you cannot get stranded inside someone else’s account.
Access rules and blocked data
Access rules control what can be scanned, not who can use the site. A rule is a blacklist or whitelist entry for either an IP (CIDR ranges work) or a URL, optionally with an expiry. URL rules match the host and all of its subdomains.
- Blacklist beats whitelist. A target matching both is refused.
- One whitelist rule turns on allowlist mode. As soon as any active whitelist rule exists, a target must match one of them or it is refused. That is the switch for a locked-down internal deployment, and it is easy to trigger by accident with a single test rule.
- It fails closed. If the rules cannot be read, scans are refused rather than allowed. A database problem must not silently disable the blocklist.
- Decisions are cached for 30 seconds per host, but hit counts still increment on a cached decision, so the panel’s counters stay honest.
Blocked Data is the cleanup half, and it is admin-only because it deletes across every user’s account. It finds scans for a domain and its subdomains, deletes them in bulk, and purges the cached public host-reputation row for a host. That last one is the takedown escape hatch: normal retention and user-initiated deletion never touch that table.
Backups and restore
The Backups tab runs the same script you would run by hand, so a scheduled run and a button press show up in one history. Only one backup runs at a time; a second request while one is in progress is refused rather than queued.
A backup is a full logical dump of the database, gzipped and then encrypted by default with AES-256-GCM. Old files are pruned past the retention window, and an optional pre-signed upload URL sends each one offsite without the application ever holding long-lived cloud credentials.
| Variable | What it does |
|---|---|
| BACKUP_DIR | Where backups are written. Defaults to ./backups. |
| BACKUP_RETENTION_DAYS | How long to keep files. Default 14. Set 0 to keep everything. |
| BACKUP_ENCRYPTION_KEY | 64 hex characters. Falls back to the API key encryption key so backups are never silently plaintext. Deliberately env-only: storing the key that decrypts the database's backups inside that database solves nothing. |
| BACKUP_OFFSITE_UPLOAD_URL | A pre-signed PUT endpoint. Works with any S3-compatible or plain HTTP receiver. |
Restoring
Restore is a command-line operation and it is destructive, so it refuses to do anything without an explicit confirmation flag and refuses to restore into a database that already has tables unless you force it. An encrypted file is fully decrypted and its authentication tag verified before a single byte reaches the database. Afterwards it prints row counts for the main tables so you have evidence the restore actually landed rather than a silent zero-row success.
Scheduled backups are off by default and run daily when enabled. Note that a database dump does not include anything stored outside the database, so back up your uploads directory and environment file separately.
Broadcast email
Broadcasts go to verified accounts only, never to unverified ones. You can target everyone, a specific plan, all paid users, or a single address, and optionally narrow further to people who have a given notification preference switched on.
Delivery walks the user list in pages with a small number of sends in flight at a time, and each recipient is claimed atomically before their email is sent. A double-click on Send, or a resend racing an in-flight send, cannot deliver twice. Failed recipients are recorded as failed so a later resend picks up exactly those.
Cleanup and retention
A cleanup job runs automatically every few minutes, pruning expired sessions, spent tokens, expired invites and each table past its retention window. Retention is a setting per table: API usage, revoked API keys, data requests, audit log, security alerts, error logs, email logs, finding feedback, notifications and cache tables each have their own.
The Settings tab also has a manual trigger behind a confirmation dialog. It is admin-only and it is not undoable: it permanently deletes scan history, audit rows, sessions and tokens that are past their windows. Use it when you have just lowered a retention value and want the effect now rather than at the next tick.
For the deployment side of running an instance, see Self-Hosting, and for the values themselves, Configuration.