Settings
All Settings in the topbar's gear menu (or Settings in the command palette) opens the settings hub. Its tiles come in six sections: Server — General, Server Information, IP Management, Plugins (§10); Hosting — PHP, FTP (§11), Spam Filter, Banned Words; Tools — Cron Jobs (§11), Migration (§8), and Python Apps (§12) and Node.js Apps (§13) while their plugins are active; Security — Security, Fail2Ban, ModSecurity WAF (§7); Panel — Notifications (§15), Whitelabel, Updates (§9), License; Integrations — API Access, Webhooks (§14). Profile opens from the account menu.
Profile
Email, language preference, avatar (≤ 2 MB, JPEG/PNG/WebP by MIME type and magic bytes; SVG and HTML are refused; the upload route allows five attempts per minute). Automatic (the default) keeps the language you sign in with; an explicit choice switches the panel to that language on every login and accepts every shipped locale, region variants and three-letter tags included (pt-BR, zh-TW, fil, …). Changing the recovery address, the language or the profile picture writes a settings_updated audit row; a save that changes nothing writes none. Password change invalidates other sessions; the new password is at least 8 characters and at most 72 bytes (UTF-8). Two-factor enrollment offers an authenticator app (TOTP) or a code e-mailed to the recovery address; the e-mail factor follows that address when you change it here, so the next code always goes to the current address, and switching 2FA off clears it. E-mail codes are sent through the server's local mail relay unless admin.two_factor.smtp_* in agent.conf names another SMTP server.
Security
- Admin IP whitelist (CIDR list or empty for any). Saving a changed non-empty list (an entry added or removed) rotates the session secret: every other admin browser session and every client session ends and has to sign in again, while the admin who saved it stays signed in with a new session cookie when their own address is on the new list (the page says so; saved with force from an address outside the list, that session ends too and the panel goes to the sign-in page), and a
secret_rotatedaudit row (Logs › Audit, filter Secret Rotated) records the rotation with its reason; the same row is written when the admin password is changed. Saving the page with the list unchanged (for example after editing only the session timeout) keeps every session. API keys are not affected. Values that equal the stored ones are dropped before anything is written, so an unchanged save writes no configuration and no audit row; a refused value answers422 VALIDATION_ERRORlike the General page (400stays for the self-lockout guard), and text where a number or a switch is expected is refused. - 2FA enforcement (off / TOTP / email / both). While the switch is on, the login and session endpoints report
requires_2fa_setup(with the demanded method) for an identity that has not enrolled a second factor, and the panels act on it: such a sign-in lands on the enrolment page (admin: Settings › Profile, client: Settings › Security) with the setup dialog open and a notice explaining why; every other panel page redirects there until the second factor is active (signing out stays possible). When the switch demands one method, the dialog offers only that method;bothleaves the choice to the user. Identities that already have a second factor are not affected. - SSH config tab — port, root login, password / key auth, allowed IPs (writes back to sshd_config; reload requires manual confirmation). The port must be free: the ports the panel itself holds (80, 443, 2000) and any port another service is already listening on are refused with
PORT_IN_USE, becausesshd -tdoes not catch a bind conflict. If sshd fails to come back on the new settings, the previous SSH configuration, firewall rule and fail2ban port are restored and the save reports a warning. - Root password — changes the Linux
rootpassword. Requires an admin browser session: HMAC API keys are refused withHMAC_FORBIDDEN_FOR_CREDENTIAL_MUTATION, like every other credential change. Control characters (line breaks) are refused. - Spam thresholds (Rspamd reject / add-header / greylist scores) are not on this page: they are set on the Spam Filter page (Settings → Spam Filter, Overview & Config tab). There too a failed read of the scores shows an error state instead of the form.
- A failed read of these settings (an error, or a rate-limit answer) shows an error state with a Retry button instead of the form, so a Save can never write the defaults, or the answer kept from an earlier visit, over the stored settings.
General
Hostname, advertised server IP, timezone, nameservers, default PHP/web server, MariaDB / FTP / mail tuning, SSL defaults, default language and the client-panel feature switches. The server IP must be a routable unicast IPv4 address (loopback, unspecified, broadcast, multicast, link-local, reserved and zero-padded forms are refused with INVALID_IP / VALIDATION_ERROR); changing it rewrites, in every zone, the A records and SPF entries that carried the old address (a DNS failure is listed in the answer's dns_errors and does not stop the change), gives the accounts that used the old address the new one, and regenerates the panel vhost; the account sites listen on every address, so their vhosts are left as they are. It does not add or remove addresses on the machine — that is IP Management. IP Management is a page of its own (/admin/ip-management, opened from the IP Management tile on the Settings page, not from General; IPv4 only): the list shows every address the host carries plus stored records the host no longer carries, marked Not bound — they are never offered to accounts. Adding an address binds it at once and persists it through netplan (Debian) or nmcli connection modify (NetworkManager, RHEL); when the persistence step fails the answer carries a warnings list: the address is bound now but will not survive a reboot. The configured primary address, the address the host's default route uses and the first address of the default interface cannot be removed, nor can an address still assigned to an account (the count is in the message). A Not bound record can still be updated and deleted; only an address that is neither bound nor stored answers 404 NOT_FOUND on update and delete. The audit rows carry the operator and their address, and changing the advertised server IP refreshes the account index the list uses.
Values sent as text where a number or a switch is expected ("10000", "yes") and fields the form does not define are refused with 422 VALIDATION_ERROR; nameservers are filled in order (an ns3 with an empty ns2 is refused instead of moving up a slot) and the MariaDB buffer sizes take a non-zero K/M/G value (0K is refused). A save that changes nothing writes no configuration, touches no service and leaves no audit row; the settings_updated row of a real change names only the changed fields (client_features.<name> for a feature switch) and carries the operator and their address. The agent's reason for a refused value reaches the toast, a 0 can be typed into the numeric fields whose minimum is 0 (Keep-Alive), and a failed read shows the error view with Retry rather than a form of defaults. Two FTP fields the API accepts, ftp_bandwidth_limit and ftp_passive_mode, are stored only: the save keeps and echoes them, but no Pure-FTPd directive is written for them and the form does not show them.
File Manager upload limit. The File Manager section sets the largest file one upload may carry, in MB (1–10240, default 256; file_upload_max_mb in the API, files.max_upload_mb in agent.conf). It applies to the File Manager in the admin and the client panel: the panel refuses a larger file when it is added, before anything is sent, and on the two File Manager upload routes (/api/v1/accounts/<username>/files/upload, /api/v1/client/files/upload) the agent and the panel's nginx take a body of up to the limit plus 1 MB. Saving a new value rewrites the panel's nginx location snippet and reloads nginx; if nginx -t refuses the snippet, the answer carries a warning and nginx keeps the previous limit. The other upload routes (application archives, migration, avatar, feedback) keep their 300 MB ceiling. An upload travels in a single request: while it runs, the server needs free disk space of about two to three times the file size (nginx's request buffer, the agent's temporary copy and the file being written), and an interrupted upload starts over — use FTP or SFTP for very large files.
Server name certificate. A section of its own on General (/admin/settings/general#hostname-certificate; the setup guide's item opens it in a new tab) shows the certificate the panel (nginx / Apache), mail (Postfix, Dovecot) and FTP (Pure-FTPd) serve on the server name: who signed it, the names it covers, its end date and its state (Valid, Self-signed, Expired, For another name). Get a certificate (Get it again once one is installed) asks Let's Encrypt for one; Let's Encrypt checks the name over port 80 through the panel's own challenge folder, so the name's A record must point at this server and port 80 must be reachable from the internet. A name that cannot carry a public certificate — no hostname yet, an IP address, a name without a dot, .test, .local, .localhost and the other special-use names — is explained on the card and the button stays off (422 HOSTNAME_CERT_NOT_ISSUABLE); a name that does not resolve answers 422 HOSTNAME_CERT_DNS_MISSING. Every configuration that names the certificate is tested before the reload, and a refused test or a failed reload puts the previous files back (500 HOSTNAME_CERT_INSTALL_FAILED); a refusal by Let's Encrypt (502 HOSTNAME_CERT_ISSUE_FAILED) changes nothing either — the error notice and the card's last attempt then show the reason it gave (what the certificate authority reported for each name, or certbot's own error line). One request runs at a time (409 HOSTNAME_CERT_IN_PROGRESS). certbot's timer renews the certificate and its hook installs the renewed files the same way. Before the first Let's Encrypt certificate is installed, the pair in place — the installer's self-signed one, or a certificate you installed (a Cloudflare Origin CA certificate, for example: Get it again replaces it) — is kept in /etc/whost/ssl/fallback/; if the Let's Encrypt certificate expires without being renewed, the agent puts that pair back on the panel, mail and FTP within 15 minutes and notifies you (Server name certificate expired, audit row ssl_replaced). With ssl.staging: true in agent.conf the requests go to Let's Encrypt's test service, whose certificates browsers do not trust; the card says so. Endpoints used: GET / POST /system/hostname/certificate.
Server Information
Server Information in the topbar's gear menu (second to last, right above All Settings), the Server Information tile in the hub's Server section or the command palette opens /admin/server-info: the whole machine on one page. It is for the administrator only — the client panel has no counterpart, and the three endpoints below refuse a client session with 401.
- Overview — host name, operating system and kernel, platform (bare metal, or a virtual machine and its type), manufacturer, motherboard, BIOS / firmware with its date, uptime and boot time, time zone and clock synchronisation, the panel's IP address and web server, the security module (SELinux mode, or AppArmor with its enforcing profiles). The dashboard keeps the CPU, memory, disk and swap gauges; this page has none.
- Processor — model with sockets / cores / threads, and right beneath it the live load (1 / 5 / 15 minutes and the run queue); then vendor, socket, base, maximum and current frequency, caches, family / model / stepping, microcode, virtualisation and hypervisor, address sizes, NUMA nodes, frequency scaling driver and governor; the CPU time split, each CPU's load, context switches and interrupts per second; the instruction set extensions and the kernel's CPU vulnerability report (vulnerable / mitigated / not affected, with details). On a wide screen, where Processor and Memory stand side by side, the Processor card takes the Memory card's height and scrolls inside under its fixed header; on a narrow screen both cards keep their full height.
- Memory — installed size and what is available now, maximum capacity and slots in use, error correction, the live split (used, buffers / cache, free, shared, kernel slab, huge pages), every memory module from the firmware's DMI table (slot, size, type, speed, manufacturer, part number, serial number, rank / voltage; empty slots listed), and swap devices with their priority and the swap-in / swap-out rates.
- Disks — each physical disk with model, size and type: the medium (NVMe SSD, SSD, or HDD with its spindle speed when
smartctlreports one) and the bus (SATA, SAS, SCSI, USB, VirtIO…; taken fromlsblk, or from udev's database wherelsblknames none, as for a virtio-scsi disk). A virtual disk shows the medium its hypervisor declares (non-rotating → SSD, rotating → HDD), marked Virtual and As the hypervisor reports it — the storage underneath may differ. Then serial number, firmware, sector sizes, I/O scheduler and partitions (size, file system, mount point, the RAID array or volume using it). SMART health is read withsmartctlwhensmartmontoolsis installed (a sleeping disk is not woken; it reads Asleep, not woken); without it an NVMe disk is read from its controller's own health log (temperature, power-on time, power cycles, wear, spare capacity, data written and read, media errors, unsafe shutdowns, error log entries), and SATA / SAS disks show a note that their health needssmartmontools. The installer does not installsmartmontools. - File systems (usage, free space, inodes, read-only marks), disk activity (read and write rates, IOPS, busy share) and RAID arrays (software RAID: healthy or degraded, active / failed / spare members, resync or rebuild progress).
- Network — interfaces (kind, addresses, link state, speed and duplex, driver and hardware, live traffic, totals, errors and drops), the default routes and IPv4 / IPv6 gateways, DNS servers (behind a local
systemd-resolved, its upstream servers) and search domains, TCP connection states, and every listening port with its protocol, address (all interfaces or local only) and process. On a wide screen, where Routing and DNS and Listening ports stand side by side, the Listening ports card takes the Routing and DNS card's height and scrolls inside under its fixed header; on a narrow screen the port list scrolls in its own box. - Processes — the process table (PID, user, command line, state, CPU, memory, threads, start time; busiest first, searchable by name, command, user or PID, filterable by state) with the total, thread and kernel-thread counts. CPU is the share of the whole machine (all CPUs = 100%) since the previous reading; at most 3,000 processes are listed.
- Software versions — the panel and the installed stack (web servers, PHP versions with the default marked, databases, DNS, mail, FTP, security, application runtimes, system), read from the package database in one query. Each row names the component with its version on the right and the package with its package version beneath. The card spans the page's full width; its categories stand in two columns once the card is wide enough (a category never splits), in one column on a narrow screen.
- PCI devices with their driver and, beside them on a wide screen, health: failed systemd units, pressure stall information (CPU, memory, I/O; some and full over 10 / 60 / 300 seconds), logged-in users and open file handles. On a wide screen the PCI devices card takes the Health card's height and its device list scrolls inside under its fixed header; on a narrow screen the list scrolls in its own box. The page shows no hardware sensor table; the agent's live answer still carries the sensor readings (temperatures and fans with their limits), and Download JSON saves them. A disk's temperature stays in its SMART health on the Disks card.
The agent collects the hardware and software inventory once and keeps it: a visit more than ten minutes later shows the kept answer, marked as being read again, while a new collection runs; Refresh waits for a new one. Download JSON, left of Refresh, saves what the page holds — the inventory, the latest live readings and the process table, exactly as the agent answered them, with exported_at and the server name — as one file named server-info-<server>-<YYYYMMDD-HHMM>.json; the browser builds it from the page's data (no extra request), and the button stays disabled until the inventory has arrived. The process command lines are in the file; the page itself never stores them in the browser tab. While the page is open the live readings are refreshed every 5 seconds and the process table every 10 seconds; rates are measured between two readings. On a 2-vCPU server an open page adds about 2% of one CPU core to the agent's load. The machine identifiers the licence's hardware fingerprint uses (product UUID, machine-id, boot ID) are not shown.
Endpoints used: GET /system/server-info (?refresh=true waits for a new collection), GET /system/server-info/live, GET /system/server-info/processes.
API Keys
Create new key (with optional IP allowlist), revoke, delete, view access logs (last 100 calls per key with timestamp, IP, endpoint, status). The secret is shown once at creation time — the operator must copy it immediately. Default API Key is the installer's pair from agent.conf; revoking or deleting it retires it for good. A key always carries at least one scope: creating or updating a key with an empty scope list answers 422 VALIDATION_ERROR. A Full Access key, the Default API Key included, may also change an account's password (PATCH /accounts/{username}/password — the account's panel login, its system user and its default FTP user; for a billing system's Change password); an Audit Read-Only key may not (403 SCOPE_DENIED). Each change is written to the Audit log as Password Changed with the key as actor.
Reseller keys. A reseller whose ACL plan grants API Access mints its own keys from the client panel (Settings → API Access). Such a key is bound to the reseller account: the Owner column names it (server-wide keys show Server), its scope is fixed to Reseller (client API) and cannot be widened, and every request signed with it is served as that reseller on /api/v1/client/* — the same ownership checks and ACL plan as the reseller's browser session, never the administrator API, whatever session cookie the request also carries. Switching API Access off in the plan refuses both minting and every request signed with an existing key; suspending the account does the same. The operator sees, revokes and deletes reseller keys here like any other key; a reseller holds at most 10.
Client sign-in links. A server-wide key with full access can create a one-time link that signs a customer into the account's client panel without the password (POST /accounts/{username}/login-token, for a billing system's "log in to the panel" button; see the API reference, Client sign-in links). The link is used once, expires after at most five minutes, can be bound to the customer's address and skips the account's second factor; the session is the customer's own and ends the account's other client sessions. The Audit log records each link as Client Sign-in Link Created and its use as a client Login Success (method login_link). The link carries its token after # (/api/v1/auth/client/login-link#token=…), which the browser never sends, so the token stays out of the web server's and Cloudflare's logs; the page it opens posts the token. The earlier ?token= form still signs in and is deprecated (WHost no longer issues it). Log in as client opens the client panel the same way.
Whitelabel
Branding overrides: company name, logo (light + dark), favicon, color theme, the background video of the sign-in pages. Gated behind the whitelabel license addon — all of them, the company name, logos and colours included; the addon is sold separately and is part of no license plan. The colours, the company name and the address of each logo are propagated to the static frontend via /var/www/whost/panel/config.json so the next page load picks them up without rebuilding; the logo images themselves are separate files (below), so config.json stays a few hundred bytes whatever the logos weigh.
The brand follows the addon: when a licence check finds it granted or withdrawn, config.json is rewritten at once, so the next page load shows or drops the brand without an agent restart; the account pages the agent writes (a new account's default page, the suspended page) take the published colours only while the addon is licensed and the feature is on. Purchase Addon and Renew Now open the WHost product page (https://wisecp.com/whost). With the feature on and hide_login_branding set (the default), the sign-in screens carry the company logo and name — the administrator form's subtitle names the company too; each screen paints with its form at once, and the logo and name appear as the page finishes loading, never the WHost defaults first. The setting is the Brand the login pages switch at the top of the Login Page Video section; switched off, the sign-in screens keep the WHost logo and name while the panels stay branded. An addon the licence service still reports as active but whose expiry date has passed is treated as expired by the panel itself: the page shows the expired banner with Renew Now, and the brand leaves the published pages at the next licence check (the publisher normally suspends a lapsed addon first; this is the panel's own safeguard).
The colour theme paints both panels in both themes, and the admin and client sidebars share one look. In the light theme the sidebar colour tints both sidebars: they stand on a dark ground mixed from it, and the menu text is a pale tone of the same hue, so it stays readable whatever the colour (in the dark theme the text takes its tone from the primary colour). The page that is open is marked in the secondary colour: a light wash behind its entry and a short bar at the sidebar's edge. The dark sidebar colour gives its hue and saturation to the dark theme's page background, cards, muted surfaces and lines, so the page reads as one scale; in the dark theme both sidebars stand on that same card surface, slightly see-through over a blur. Those surfaces stay dark enough for the panel's secondary text to keep a 4.5:1 contrast: with a lighter dark sidebar colour (for example one picked up from a mid-tone logo) they take a darker step of the same hue instead of the sidebar's own lightness. The primary and secondary colours draw a short glow behind the logo in the top corner of both sidebars (top right in right-to-left languages); it takes the light-theme pair in both themes and looks the same in each. Text on the primary and secondary colours (buttons, badges, tooltips) stays white while it keeps a 4.5:1 contrast; on a lighter brand colour it turns to a dark shade of the same hue.
The dark surfaces — the sidebar and its collapsed icon — show the dark logo and the dark favicon; when only the light variant is uploaded, the light one stands in (inside a light frame when it is too dark for the sidebar), and the WHost mark appears there only when neither variant is uploaded. When no favicon is uploaded at all, the browser tab and the collapsed sidebar show a generated mark instead of the WHost icon: the first letter of the company name (of the server name when no company name is set) on the primary colour. If the page cannot read the addon status or the settings (a limiter answer, a network error), it shows an error with Refresh instead of the purchase banner and a locked form; the dashboard's purchase prompt likewise opens only on a real "not licensed" answer.
What an uploaded logo may contain. The file type is decided by the leading bytes, not by the declared content type: JPEG, PNG, WebP and SVG are accepted, anything else is refused with 415, and the ceiling is 2 MB. An SVG is rewritten before it is stored — XML entities are refused outright (no file:// reads, no expansion bombs), and <script>, <foreignObject>, on* handlers, javascript: links, animations that rewrite a link target (under any prefix) or carry a script or outside address in any of their values, stylesheets and style or presentation attributes that pull a second document (read as a browser reads CSS, escapes decoded — only a url(#…) pointing inside the file stays), xml:base, and links to anything other than a part of the same file or an embedded PNG / JPEG / GIF / WebP image are removed. Ordinary shapes, animation and styling survive. An SVG may open with an XML declaration, a comment of any length or a doctype — the root element decides the type.
How a logo is stored. A JPEG, PNG or WebP is shaped before it is stored: a logo is scaled down to fit 512 px wide by 160 px tall — the panel shows a logo at most 160 px wide and 48 px tall, and this covers the sharpest screens (proportions kept, a smaller image is never enlarged, the format stays the same), and a favicon becomes a square PNG of at most 256 px — a non-square image is centred on a transparent square, nothing is cropped. The camera orientation of a photo is applied. A small image that already fits (up to 256 KB) keeps its pixels byte for byte; an animated logo that fits keeps its animation, a larger one keeps its first frame. What a camera or an editor wrote beside the pixels (EXIF, XMP, text comments: place, device, date) is removed from every stored image, logos and sign-in covers alike, since the sign-in page shows them to anyone; the colour profile stays. An image above 25 megapixels is refused from its header with 400 IMAGE_DIMENSIONS_TOO_LARGE before anything is decoded, and a damaged or truncated file with 400 INVALID_IMAGE. The upload answer reports each file's size before and after (optimized). SVG is kept as a vector. The stored file is served by the panel's web server at /branding/<name>; the name changes whenever the image changes, so browsers keep the file for a year and pick up a new logo at the next page load. Replacing or removing a logo, and the reset, delete the old file. The files live under /var/lib/whost/branding/ on the server — include that folder in host backups.
Login page video. The Login Page Video section replaces the background video of the administrator and of the client sign-in page, one card each (the badge reads Default or Custom; the preview plays on hover or click). Upload Video takes an MP4, MOV (as phones and cameras record), WebM or MKV file and prepares it in the browser before it is sent; whatever it came in, the stored file is an MP4. An H.264 video that fits 5 MB is only repackaged (a MOV or MKV becomes an MP4 without re-encoding). One in another codec that fits (HEVC from a phone, VP9 from a WebM, for example) is converted to H.264 at about its own picture quality — at most three times its bitrate, as H.264 needs that much more for the same picture — and is never grown to fill the 5 MB; the card reads Converting the video and the notice says it was converted to MP4. A larger one is re-encoded as H.264 — scaled into 1920 × 1080 (1080 × 1920 for an upright video, the phone's rotation applied), at most 30 frames a second, aimed close to 5 MB: a browser's encoder often stays well under the bitrate it is given (or runs over it), so a pass that lands under about 4.25 MB or over 5 MB is encoded again with the difference it measured, and the card reads Fine-tuning the video during that pass. The encoded width is a multiple of 16 (a hardware encoder given a width that is not a multiple of four loses most of the picture's detail), and a key frame is written every 10 seconds, which leaves more of the 5 MB to the picture. Within 5 MB a clip of 10–20 seconds keeps most of its detail; a long one, or one with fast motion and fine texture everywhere, loses visibly — a shorter clip of the same scene looks better than the whole of a long one. A video so long that even 640 × 360 would drop below about 300 kbit/s is cut to the length that fits, and the panel says how many seconds it kept. Either way the sound track and the file's descriptive tags (camera, place, date) are dropped — the sign-in page plays muted. The cover the page shows until the video plays is taken from the video's first frame, so the picture does not jump when playback starts; Change Cover replaces it with a JPEG, PNG or WebP of your own (scaled into the same box in the browser and again on the server, 2 MB at most), and a new video brings its own first-frame cover again. Restore Default removes the video and its cover and the page shows the default video. Compression needs a browser with WebCodecs (current Chrome, Edge, Firefox and Safari); elsewhere an MP4 of at most 5 MB is uploaded as it is, without a cover, and a larger one, or a MOV, WebM or MKV, is refused. The agent stores the video as it arrives, after reading its structure (no frame is decoded on the server), with its descriptive tags, XMP packet and creation times blanked — an MP4 uploaded as it is loses them on the server too: it must be an MP4 whose video track is H.264 (415 UNSUPPORTED_MEDIA_TYPE names the codec otherwise) of at most 5 MB (400 FILE_TOO_LARGE); a damaged file answers 400 INVALID_VIDEO, a cover sent for a page that has no video of its own 400 LOGIN_VIDEO_REQUIRED, and a new video sent without a cover drops the old cover. The video and the cover are files under /branding/ like the logos and follow the addon and the feature switch the same way: when the addon lapses or whitelabel is turned off, both sign-in pages play the default video at the next page load. A sign-in page loads its video after the panel script has read the branding, so a host with its own video never downloads the default one; until then the cover is shown.
Every change — a settings save, a logo upload, a logo removal, a login page video upload or removal, a reset — writes a settings_updated audit row (scope: whitelabel); a save that changes nothing writes none. Colours are HSL triples within range (hue 0–360, saturation and lightness 0–100) and the two switches are booleans ("yes" is refused). Removing a logo that is not set answers 404 LOGO_NOT_FOUND, removing a login page video that is not set 404 LOGIN_VIDEO_NOT_FOUND. While the addon is unlicensed or the feature is off, the form is disabled for the keyboard as well as the mouse; while the addon is unlicensed the agent refuses the settings save, a logo or login page video upload or removal and the reset with 403 LICENSE_ADDON_REQUIRED, so nothing the panel will not publish can be stored (the values already stored stay as they are and the public branding read stays inactive). A limiter answer on any save is reported as an error, and the colour fields fall back to the stored values when their auto-save is refused.
Audit Logs
Every authentication, mutation, and admin action lands in /var/log/whost/audit.jsonl and is surfaced on this page with filters (actor, event, time range). Client-side identity events are included: client login/logout and 2FA challenges, plus password changes (including resets through the mailed link), 2FA enable/disable, backup-code regeneration and recovery-email changes made from either panel. A refused sign-in is a login_failed row for the admin and the client form alike (the row's details name the reason — unknown identifier, wrong password, wrong second-factor code — and the form). When the attempt counter closes a sign-in form or a 2FA step for a while, a login_locked or 2fa_locked row records it (admin and client forms alike, with the lock length in the row's details). The server's own state is here too: a license_state_changed row for every licence transition (both states and the trigger in the details — the row never carries the key), an integrity_alert row when a self-check of the running agent found its code or manifest changed (written before the agent stops itself, naming the check) or when the start-up manifest check refused to start it, and the update_install_started / update_installed / update_failed rows of a software update. Append-only; past 10,000 entries the oldest are moved to audit.jsonl.1 beside it (an archive past 50 MB becomes .2, replacing the previous one) and the live file keeps the newest 8,000 — the page reads the live file.
Banned Words
Wordlists of banned terms. The lists are enforced: an account name, a primary domain, an addon domain or a subdomain that carries a banned word is refused with 403 BANNED_WORD (the answer names the word and the list) when it is created or changed. Matching is case-insensitive and by substring: the value and the parts it splits into at ., -, _, / and @ are compared with every word of every enabled list, so malware also matches antimalware-tools — keep the lists' words specific. Eight built-in lists ship with the panel (five enabled); a built-in list can be switched off and edited but not deleted (422). The Check Value box asks the same check the gates run. A list file under /etc/whost/wordlists/ that cannot be read is named in a notice on the page and is not applied.
Endpoints used: GET/POST /banned-words, GET/PUT/DELETE /banned-words/{wordlist_id}, POST /banned-words/check.
Our support team is here around the clock for anything you can't find above.