Service → LibreNMS bill mappings. Preview computes the current/previous overage; Run (dry) previews the WHMCS push; Run (charge) pushes it.

Map a WHMCS service to a LibreNMS traffic bill (adopt an existing bill or auto-create one). Pick the client, the service, and the interface(s) — no id typing.

Read the previous (closed) LibreNMS period, compute overage, and (unless dry-run) push it to WHMCS. Idempotent per (service, period).

WHMCS service → per-token LLM billing. Grant trial credit, preview the period, and run billing (idempotent per period). Usage is metered from the inference proxy.

Map a WHMCS service to per-token pricing. Grant a trial credit ($ of free usage) to attract the client; enable "enforce credit limit" so a trial can't overrun before they add a payment method.

Aggregate the previous (closed) period's tokens, apply trial credit, and (unless dry-run) push the net overage to WHMCS. Idempotent per (service, period).

Adopt an existing hand-built PVE cluster, then add or remove nodes. Distinct from Clusters above, which is the read-only NetBox mirror — this drives the real lifecycle. Every action previews first; nothing is applied until you confirm.

NetBox virtualization clusters (from the read-only PVE mirror) with node + VM counts. Click a cluster to drill into its member hypervisors and every VM — current host, vCPU/RAM/disk, storage backend, CPU vendor, HA, and backup coverage.

Register Proxmox endpoints (a standalone node or a per-site cluster). Nodes can't be clustered across sites, so each location is its own endpoint. The API token secret is encrypted at rest (never shown). Test probes /version + node-online live.

Clone a golden template on a registered node, optionally attach whole GPUs by PCI bus address, put the service NIC on a subnet, claim one host IP, cloud-init it. GPUs optional — serves GPU and plain VMs alike. Runs as a background job.

Look up a VM's NetBox record + live Proxmox status, or tear it down (destroy the Proxmox VM + free its NetBox IPs). Deprovision requires typed confirmation.

Live GPU pool (passthrough-mapped, free/assigned) + last burn-in result per GPU. Sync reflects the pool into NetBox inventory items. Burn-in provisions a throwaway 1-GPU VM, runs gpu-burn, scores PASS/FAIL, and deprovisions — sequential per node.

Does every VM's VLAN reach the node it runs on — and every node it can migrate or fail over to? A tagged NIC on a bond-backed bridge needs its VLAN on each node's trunk. Broken = missing on its current node; breaks on migration = missing on another node (the Smart Proxy VM was migrated to a node without VLAN 2002 and all PXE at the site stopped). Read-only.

Configure a switch port for Foreman bare metal provisioning. This looks up the server in NetBox, traces the cable to find the connected switch port, saves the current configuration, and sets the port to access mode with the Foreman VLAN.

Restore a switch port to its original configuration after provisioning is complete. This finds the backup, restores the original configuration, and marks the backup as restored.

Assign one or more VLANs from the site's customer VLAN group to a server's bond trunk and (optionally) build the switch. One VLAN may be native; each VLAN may auto-allocate a subnet, reuse an existing one, or be L2-only. Limited to 5 VLANs (QinQ beyond that).

Create ONE private (RFC1918/VRF, L2-only) VLAN and assign it to several selected servers in one call — e.g. a Proxmox cluster's private network. All nodes must share a switch (cross-switch selections are refused — inter-switch trunking to the private core is a later phase). Each node's bond trunk is rebuilt as a background job.

The ONLY endpoint that changes which VLAN is native (untagged/management) on a server's bond0 — assign/cluster/attach never touch the native. The target VLAN must already be a bond0 member. Demotes the old native to tagged, promotes the target, and rebuilds the switch. Changing the native can interrupt management access (IPMI-level recovery if it goes wrong) — use carefully.

One screen to onboard a hand-racked server: model the bond (eth0/eth1→bond0), stamp NIC MACs from the BMC over Redfish (no PXE needed) with a 10G confirmation check, record the IPMI address + admin credentials (ARPHost-Private VRF + custom fields), and reflect the live switch trunk's VLANs into NetBox. Optionally auto-create the cabling from the rack standard. Always dry-run first — the preview makes zero writes.

Release/delete customer VLANs from a server's bond: removes membership, frees IPs/prefixes, deletes the VLAN records, and tears down the switch (reverts the ae to LACP-only when the last VLAN is removed). Use this instead of admin IP release for bonded VLAN devices.

Add one address family's subnet (prefix + gateway + server host IP + Network/Broadcast) to an existing VLAN on a server's bond, then build the irb gateway on the switch.

Delete just one address family's subnet (host IPs + gateway + prefix) from a VLAN on a server's bond — keeps the VLAN + the other family — and drops irb.{vid} family inet/inet6 on the switch.

Create a tenant-owned shared VLAN + public subnet that multiple servers can attach to. No server is attached yet — use Attach afterward.

Allocate a v6 /64 (gateway ::1) for an existing shared VLAN and rebuild attached servers' switch so irb.{vid} gets its v6 gateway. Use the vlan_netbox_id (not the VID). 400 if it already has IPv6.

Add a server to the shared VLAN — claims the next host IP from the shared subnet and builds that server's trunk.

Remove a server from the shared VLAN — frees just that server's IP and rebuilds its trunk. The VLAN+subnet is deleted only when the last server detaches.

List all port configuration backups with optional filtering by status and device.

Get detailed information about a specific configuration backup.

Delete a backup record from the database. This does NOT restore the port configuration.

Query the current power state of a server via IPMI/BMC. Returns whether the chassis is powered on, off, or unknown.

Query server hardware information via IPMI FRU data. This is a read-only out-of-band query that works even when the server is powered off. Returns basic hardware info as reported by the BMC (varies by vendor). Does not update NetBox.

Find inventory that needs a WHMCS product or pricing review. Includes reserved and unready servers. A mapped product does not establish ordering readiness.

Provision jobs that ended cancelled or failed, but whose device NetBox shows as actively in service anyway — the server installed and went live while its job was closed out as unsuccessful, so nothing flipped the WHMCS service to Active. Derived, never stored.

List servers available for sale from inventory. Only returns servers with status "inventory" in NetBox, "available-for-sale" tag, that have been wiped and passed hardware testing.

Paid self-serve orders awaiting provisioning. Preview shows exactly what the build will do, from the configuration the customer bought — then Approve provisions it. The order's stored details (OS, hostname, SSH key, IP sizes, bandwidth) drive the build; nothing is re-entered here.

Orders placed in the WHMCS cart reach core-api through the server module's CreateAccount. Each one is listed here until it becomes a store order (it then appears in the Fulfillment Queue). Review and Failed rows need a person: fix the cause named in the row, then Retry to run every check again from the start. Nothing is reserved or built for these rows.

Active self-serve holds. Each row locks a device (a table-only hold — the device stays at NetBox status "inventory") until it is assigned to a client or released. This list also feeds the Assign and Release selectors.

Place an admin hold on an available server (no WHMCS order). The hold is a table row — the device stays "inventory" but is excluded from availability until released or assigned.

Permanently assign a server to a WHMCS client/service: stamps the device tenant + WHMCS ids, sets NetBox status to "active", and deletes any outstanding reservation. Requires the typed confirmation ASSIGN-{device_label}.

Cancel a hold — deletes the reservation row so the device is available for sale again (no NetBox status change).

WHMCS decides what is sold; Foreman decides what can be installed. This reconciles the two: a customer is only ever offered an OS that exists in both. Review the suggestions, tick the ones that are right, and approve them — nothing becomes sellable until you do.

Release a server back to the available inventory pool. This performs: 1) Disables all switch ports (physical + bond/ae interfaces), 2) Powers off the server (hard power off via IPMI), 3) Updates NetBox device status to "inventory".

Boot server into discovery mode to collect hardware inventory. The server will PXE boot into a discovery image that collects CPU, memory, storage, network interfaces, and IPMI/BMC details.

Look up an active discovery job by server MAC address. Used internally by hardware-discovery.sh on FDI when kernel cmdline parameters are not available. Returns the callback_url and job_id for the discovery script to use.

List all operating systems available in Foreman for provisioning. Use the id or title from the response as the os_override parameter when provisioning.

List all host groups available in Foreman for provisioning. Use the name or title from the response as the hostgroup parameter when provisioning.

Get comprehensive status from all integrated systems: NetBox (device info, lifecycle state, IPs, interfaces), Foreman (build status, host group), WHMCS (service status, client info), IPMI (current power state), and active jobs.

Provision a server with an operating system. The server must have NetBox status "inventory" to prevent accidental reinstallation of production servers. This workflow configures switch ports, sets up Foreman, PXE boots the server, and waits for the installation callback.

Securely wipe all disks on a server. Boots server into wipe environment and securely erases all disks. After successful wipe, hardware testing is automatically triggered unless skip_test=true.

Run hardware tests on a server. Boots into FDI test environment and runs CPU stress test (stress-ng), memory test (memtester), SMART health check, disk I/O benchmark (fio, full/burn-in only), and network link verification (ethtool). On success, server is automatically returned to inventory unless auto_return_to_inventory=false.

Rewrite a server's storage controller. Pick a layout per drive set, preview the exact storcli commands, then apply. The server PXE-boots to do the work and reports back.

Finds servers recorded with a storage controller they do not have. An older discovery script counted storcli's “No Controller found” reply as a controller, which left an Unknown controller item and has_hardware_raid=true behind — and a “Storage controller” on the store listing. The server judges each device against its own stored storcli capture; this page only shows that verdict.

Run the audit first: it is a dry run over every device that claims a controller and changes nothing. Its rows are the selector — Apply is offered only on a row the API classified proven_phantom. A server that can be PXE-booted is better fixed by re-running discovery.

List all baremetal workflow jobs with optional filtering by status, workflow type, and device.

Get detailed information about a specific baremetal job including step-by-step logs and progress.

Cancel a pending or running baremetal job. Only jobs with status "pending" or "running" can be cancelled.

Purge completed job history to remove sensitive data like passwords. By default, only purges jobs with status "completed", "failed", or "cancelled". Use include_all=true to purge ALL jobs including pending/running ones. This is a destructive operation - purged jobs cannot be recovered.

Manually send a wipe completion callback for testing. This simulates what hardware-wipe.sh sends when disk wipe completes. Use leave_powered_on=true to prevent the server from being powered off after callback processing.

Manually send a discovery completion callback for testing. This simulates what hardware-discovery.sh sends when hardware discovery completes. Use leave_powered_on=true to prevent the server from being powered off after callback processing.

Allocate IPv4 AND IPv6 addresses from site's prefix pools. Creates a dedicated IPv4 prefix (/31 point-to-point up to a /27 block) plus an IPv6 /64, with the network, gateway, server and broadcast IPs populated. Pool selection is automatic: primary pool for new servers, secondary for additional IPs.

Assign one or more caller-specified public IPv4 blocks (exact network-aligned /27–/31 CIDRs) to a routed/bonded server — the primary block sets the device's primary IP, each secondary block's gateway is added to the switch ae so its source range is permitted. Every prefix + IP is tenant-owned. IPv4-only for now.

Assign a specific (already allocated) IP to a device interface. Use for manual IP assignment or moving IPs between servers.

Release a server's IP allocation back to the pool. Deletes IP addresses assigned to the device and clears primary IP references.

Change which IP is the device's primary. Updates NetBox primary IP pointer and all related descriptions. No switch changes.

Route a secondary subnet to a next-hop IP via static route. Removes gateway from switch interface, adds static route and prefix-list entry. All IPs in the subnet become client-manageable.

Restore a statically routed subnet back to interface mode. Re-adds gateway IP to switch interface, removes static route.

Reconcile the switch's static routes against NetBox and classify each into buckets: managed (in sync / drift), orphaned on switch, missing on switch, customer BYOIP, customer BGP, ARPHost infra, and unknown-needs-review. Read-only. Provide exactly ONE scope.

Compare every device's BMC address (oob_ip) with the rack-unit rule rack IPMI prefix network + 1 + RU. Drift is listed first. Read-only: it never renumbers a BMC.

Delete static routes classified orphaned_on_switch (route + prefix-list member on a switch with no NetBox record). Each named subnet is re-verified orphaned server-side (a live/managed/BYOIP/infra subnet is refused). Best populated from the Audit above (select orphans → "Send to Cleanup").

Register customer-owned IP space (BYOIP): its origin ASN, whether they run their own eBGP session, and the owning tenant. Mapping-only + conflict-safe. Admin-entered until announcement-authorisation validation exists.

Tag an ARPHost-owned facility aggregate (e.g. the supernet a site announces) so the reconcile classifies routes inside it as ARPHost infra rather than unknown. Optionally tenant-owned by ARPHost.

Audit + non-destructive migration: adopt legacy - routed X next-hop Y description markers into the authoritative route_next_hop custom field. Descriptions are never modified. The report IS the audit — run dry_run first.

List all users in the system with their API key counts. Requires admin privileges and JWT authentication.

Get detailed information about a specific user including their API keys. Requires admin privileges.

Update a user's account settings such as active status and admin privileges. Note: You cannot revoke your own admin status or deactivate your own account.

List all API keys for a specific user. Requires admin privileges.

Update an API key's settings including active status and scopes.

Permanently delete an API key for any user. This cannot be undone.

Create a new user account. The user can be assigned to a tenant and given admin privileges.

Change any user's password. The user will need to use the new password on their next login.

Create an API key for any user. The full key is shown only once!

Generate a temporary token to act as another user. The token expires in 1 hour and cannot be refreshed. Cannot impersonate yourself or other admins.

View audit logs of admin actions. Logs auto-delete after 60 days.

Synchronize tenants and users from NetBox. This imports tenants and their contacts as users, creating API keys with appropriate default scopes.

Pull active WHMCS clients and create/update NetBox tenants (with the whmcs_client_id custom field). Idempotent; never deletes. Run /tenants/sync-netbox afterward to bring them into the Core API DB.

ACTIVE WHMCS services that classify as neither server nor vm — the device sync's blind spots. The classifier is an allow-list, so a product group with no matching keyword is invisible: no NetBox device link, no tenant portal access and no bandwidth billing. Anything flagged likely_device carries an IP, so it probably has real hardware behind it.
Check this after creating a WHMCS product group, then add a substring to WHMCS_SERVER_KEYWORDS (physical device: dedicated or colo) / WHMCS_VM_KEYWORDS and re-run Sync WHMCS Devices.

Map EXISTING NetBox devices to their WHMCS client: sets device tenant + whmcs_service_id/whmcs_client_id and tenant-owns the service's existing IPs, so synced tenant users see the devices they own in WHMCS. Never creates devices or IPs; unmatched services and conflicts are reported. Run /tenants/sync-whmcs first so tenants exist.

Push IPMI address, BMC user + password and OpenVPN location into each server's WHMCS service custom fields. New provisions and reconciles write these automatically — this repairs servers built before the write-back existed.

Ownership decides overwrite, per device. A device holding ipmi_admin_* is one we built, so its BMC credentials are ours and are overwritten. A device without them is colo: the credentials belong to the customer, so they are fill-only — never generated, never replaced. Only values we actually hold are sent, so a hand-typed WHMCS field is never blanked.

Run the audit first: it is a dry run over every server, and its report is also what populates the device selector below. Devices whose WHMCS product does not define the fields appear under missing_on_product — that is a WHMCS admin fix, not a code one.

Model Active WHMCS "Virtual Servers" as NetBox virtual machines: creates a per-hypervisor cluster (from the WHMCS server_name) + the VM (tenant-owned, with its IPs), or updates an existing one to current. Creates VMs + missing IPs. Services with no hypervisor (e.g. Cloudflare) are skipped. Run /tenants/sync-whmcs first so tenants exist; use /reconcile-whmcs-vms for cancellations.

Clean up synced VMs whose WHMCS service left Active: Suspended → VM set offline (kept); Terminated/Cancelled/absent → VM deleted + its IPs freed (unassigned + tenant cleared, IP record kept). Manually-created VMs (no whmcs_service_id) are never touched. Skips on an empty WHMCS service set (safety).

Manage client-private RFC1918 space. Invariant: the global table holds only public space; all RFC1918 lives in a per-client VRF (ARPHost included). All actions are dry-run by default and never create devices.

Reconcile a duplicate tenant: reassign the source tenant's NetBox objects (devices/IPs/prefixes/VLANs) and Core API DB users to the target, then delete the source. Use the NetBox tenant IDs. Run a dry-run first.

List all reverse DNS zones configured on the PowerDNS server.

Create a new reverse DNS zone. Examples: 20.109.78.in-addr.arpa (IPv4 /24), c.d.4.f.2.0.6.2.ip6.arpa (IPv6 /32).

List all PTR records for IPs assigned to a specific device. Also shows IPs without PTR records.

The default/standard reverse-DNS record every static IP gets. A deleted PTR reverts to this (never blank), and IP assignment fills it in. Edit it here without a code deploy. Leave a field blank to revert it to the env default.