PMClient

A native console for Proxmox VE.

The cluster tree, real VNC consoles for virtual machines, text consoles for containers and node shells, and the Proxmox web interface itself — in one app, with migration, replication, backup and node updates a right-click away. Windows, macOS and Linux; x86_64 and ARM64. It follows your system theme out of the box — or pin it light or dark — and consoles stay dark regardless, as they should.

Download 1.5.4 Why it exists

Version 1.5.4 · clean-room protocol engine · 7,000+ automated tests

PMClient during a live migration: the guest's tree row reads migrating, disk 41%, while its Windows console stays connected and the guest runs a speed test at 5.5 Gbps — A guest migrating between nodes — disks copying, 41% through — while its console stays connected and the guest keeps pulling 5.5 Gbps through a speed test. The tree is filtered to one tag. *(Illustration shown if the screenshot is missing.)*

Why it exists

Running a cluster shouldn't mean juggling tabs.

Proxmox's web interface is capable, but a working day with it spreads across browser tabs, and its browser console has no seamless clipboard — so passwords get retyped and text goes through side channels.

PMClient puts that work in one app. The cluster tree, real consoles and terminals, and the web interface share one window. Migration, replication, backup and node updates are a right-click away, and the Balance page shows where the cluster is uneven.

And the console does what the browser can't: copy and paste in both directions, bound to your real system clipboard through PMClient's own clean-room RFB 3.8 engine. Turn on clipboard=vnc for the VM — PVE 8.1+, one click from inside PMClient — and Ctrl+C / Ctrl+V just works.

And when paste is dead, PMClient tells you why. It checks whether the guest's agent is actually running and reports what it found, instead of leaving you guessing at a silent clipboard.

Features

Everything between you and the guest, done properly.

Clipboard that just works

The RFB Extended Clipboard, both directions, wired to your OS clipboard — with one-click enablement on the VM and real detection of a dead guest agent instead of silent failure.

A clean-room VNC engine

RFB 3.8 written from the specification: ZRLE, Tight, Hextile and Raw decoders, resize and cursor handling, QEMU extended key events. ZRLE by default; switch to Tight for slow links.

The whole cluster, one window

Server, Folder and Pool views of the resource tree, a native datacenter summary, storage contents at a glance — and the Proxmox web UI itself, signed in already when you connected with a password or your identity provider, embedded in the pane on Windows, macOS and Linux. Where a desktop will not allow the embed, it opens in its own window and says so.

Real terminals

Text consoles for containers, node shells and serial VMs, backed by a genuine VT engine — vim, htop and other full-screen programs behave the way they should. And they survive: navigate away, come back, and the session is still there — scrollback, running command and all — until you disconnect.

Sign in your way

Passwords, TOTP second factors, API tokens — or browser sign-in through any OpenID Connect realm the cluster trusts: Entra, Okta, Keycloak, Authentik. PMClient fronts PVE's own OpenID flow, using the redirect your web UI already has.

Security without escape hatches

SHA-256 certificate pinning: confirm on first contact, loud refusal on mismatch — and deliberately no "ignore" button. Secrets are never written as cleartext — DPAPI, Keychain or Secret Service holds the key, and only an encrypted token is kept beside the profile.

Migration without guesswork

Right-click a guest: every target shows its route and cost up front — replicated, shared storage, or how much local disk gets copied. A node that can't take the guest still appears, grayed out, with the reason beside it. That covers problems Proxmox's own list doesn't flag, like a node left on an older version partway through an upgrade. If an HA rule blocks the move, PMClient names the rule and links to the HA rules page.

Blockers, named up front

If a guest can't move at all — a linked clone, a device pinned to its node — the menu says Blocked and why before anything starts, not minutes into a transfer. And unknowns fail open: a fact PMClient can't check never hides a choice you're entitled to.

Watch it move

Before committing you're told what to expect: a running VM pauses briefly, a container restarts. Then a live chip on the VM's row names the phase that dominates — migrating · disk 85%, migrating · RAM 42% — parsed from the cluster's own task log.

Consoles that follow the guest

A live migration kills the old node's console proxy. PMClient notices, waits until the guest is actually running, re-resolves its new node and reconnects with a fresh ticket — bounded backoff instead of a hung window.

Resolutions, without the console fight

Pick a resolution from the console bar and a Linux guest on a virtio-gpu or qxl display follows it — up to 4K, nothing to install. Windows guests get a one-click switch to a virtio-gpu display that unlocks the full resolution list inside the guest — and the console tracks whatever the guest picks.

Keep nodes current

Right-click a node to refresh its package index or run the full upgrade — in a real console you watch, not a spinner. A status line names what's pending by origin (17 Debian, 4 Proxmox), and when a newer kernel is installed than the one running, PMClient says a reboot is due.

Fluent in PVE permissions

PMClient reads your account's effective privileges — pool inheritance, privilege-separated API tokens and all — and shapes itself to match: power and migration actions you lack stay visible but disabled, naming the privilege you are missing. A refused ACTION is explained in plain language that names the privilege to ask for, where the client can tell which one is missing — not a raw API error.

Native on every desktop

Windows, macOS and Linux — x86_64 and ARM64 natively on each, from Snapdragon X laptops to Apple silicon.

Replication, from the right-click menu

See every replication job on a guest with its live status, run one on demand, and set up, reschedule or delete jobs without leaving the tree. Add a second target from the same menu — it lists only the nodes not yet covered, and says so when every node already is.

Back up now

Right-click a guest, pick a datastore from the ones that can actually take a backup, confirm, and watch it run. If there is nowhere to back up to, the menu says whether no datastore exists or this account simply cannot see one.

Balance: see where the cluster is uneven

A page under the datacenter showing how guests are spread across nodes, with suggested moves that would even the load and the reasoning behind each one. Suggestions only — nothing moves until you decide it should. It also shows the cluster's own scheduler settings, each with a line explaining what it does, and changes them where your account may.

In action

A closer look.

The Proxmox web UI embedded in PMClient's pane on Linux, already signed in, open on a guest's Hardware page — The Proxmox web UI, already signed in and embedded in the pane — here on Linux, opened straight to a guest's Hardware page.

A node shell in PMClient running btop full-screen in the built-in terminal — A real terminal — btop running full-screen in a node shell.

An LXC container's text console in PMClient running htop full-screen, showing the container's own services — Containers too — a full-screen program in an LXC text console, drawn by PMClient's own terminal.

PMClient's right-click menu on a running VM: power actions, and a Migrate-to submenu naming the replicated route to the other node — Right-click migrate — every route named up front.

The same Migrate-to submenu on an unreplicated VM, naming the cost before committing: the target node, how much local disk gets copied, and the storage it lands on — No shared or replicated storage? The price is on the label — how much gets copied and where it lands, said before you commit.

PMClient's tree grouped by Proxmox tag, with community-script, web and wickasoft as top-level groups — Or group the tree by tag — your Proxmox tags become the top level, so “everything tagged *web*” is one place, not a search.

A guest hover tip in PMClient showing status, node, CPU and memory, its disk, last backup date, and an hour of CPU history as a sparkline, with unchecked items named as unchecked — Hover a guest and it tells you what it knows — its disk, when it last backed up, an hour of CPU history — and names what it has *not* checked rather than leaving a gap.

PMClient on macOS showing the Datacenter status page: cluster CPU and memory, the fullest storage, a storage table, both nodes online with their uptime, and virtual machine and container counts that include a State not read line — Datacenter status: CPU, memory and the fullest storage at a glance, both nodes with their uptime, and guest counts with a *State not read* line — a guest whose run state could not be determined is counted there, not folded into running or stopped.

PMClient's Balance page showing the cluster's CRS settings, node demand, and per-guest move suggestions, with the visible suggestion buttons disabled and a tip explaining that the cluster's scheduler reverses a move that raises imbalance above its threshold — Balance ranks the moves worth making, then checks each against the cluster's own scheduler. Where auto-rebalance could move a guest back, the suggestion is shown disabled with the reason on it — as here — and the analysis still stands.

PMClient's Access pane on a guest: stored role bindings above, and below them an effective-privileges table listing each privilege with the role that grants it, the path it comes from, and whether it is applied or superseded — Manage access answers what an account can actually do here — every privilege with the role that grants it, the path it comes from, and whether it is *applied* or *superseded* by a binding nearer the guest.

The same Access pane opened by a limited account, showing three notices that name what could not be read and the privilege each would need, above a short effective-privileges table — Signed in as a limited account, the pane says what it *cannot* see and names the privilege that would show it — rather than presenting a short list as though it were the whole picture.

PMClient's Resolution menu open over a Debian guest's console, offering 1024×768 through 1440p with the guest's current 1920×1080 marked — Resize the guest from the console bar — the sizes this display will take, with the current one marked.

PMClient over a Windows 11 guest, honestly reporting that Windows ignores host-side resizes and naming the in-guest Settings path to change resolution instead — Windows won't take a host-side resize — so PMClient says so, and names the in-guest path that works. No pretending.

PMClient running natively on macOS: a Windows guest's console live beside a cluster tree filtered to the guests matching win11 — The same client, native on macOS.

Yes — that was a live migration without shared storage. A running VM whose disks live on plain local storage — ZFS, LVM, a directory — can still change nodes: Proxmox mirrors the disks to the target while the guest keeps working, then switches over in a brief pause. No SAN, no Ceph, no replication job required. The fine print is short: attached disks only, no local ISO left in the CD drive, no passthrough hardware, and snapshots on local disks only move when replication carries them — and PMClient says Blocked, with the reason, before anything starts. The home-page shot was taken mid-copy, benchmark still running inside the guest.

Download

Version 1.5.4

Pick the build that matches your OS and CPU. Public downloads are being staged — the links below lead to a holding page until they open.

OS / CPU	File	Notes
Windows · x86_64	`WickaSoft-PMClient-1.5.4-win-x64.exe`	The normal choice on desktops and most laptops.
Windows · ARM64	`WickaSoft-PMClient-1.5.4-win-arm64.exe`	Snapdragon X, Surface Pro X/11. The x86_64 build runs under emulation, but native is much faster.
Linux · x86_64	`wickasoft-pmclient_1.5.4_amd64.deb`	Debian / Ubuntu package: `sudo apt install ./wickasoft-pmclient_1.5.4_amd64.deb`
Linux · ARM64	`wickasoft-pmclient_1.5.4_arm64.deb`	As above.
Linux · other distros	`…-linux-x64.tar.gz` · `…-linux-arm64.tar.gz`	Untar and run — the tarball preserves the executable bit.
macOS · Apple silicon	`WickaSoft-PMClient-1.5.4-osx-arm64.zip`	A regular .app bundle — unzip and drag to Applications.
macOS · Intel	`WickaSoft-PMClient-1.5.4-osx-x64.zip`	As above.

Not sure which CPU you have? Windows: Settings → System → About → System type. Linux and macOS: uname -m.
Setup steps, shortcuts, permissions and troubleshooting live in the documentation.

What you need on the cluster: Proxmox VE 8.x or 9.x (tested mostly on 9.2), API on the default port 8006. Seamless clipboard needs PVE 8.1+ with clipboard=vnc on the VM's display — PMClient can flip that for you — plus the clipboard agent in the guest: part of the virtio-win guest tools on Windows, the spice-vdagent package alongside the usual qemu-guest-agent on Linux.

OpenID sign-in (Entra, Okta, Keycloak, …) opens the provider's page in an embedded browser on Windows (WebView2, preinstalled on Windows 11) and macOS (WKWebView, nothing to install); on Linux it appears in its own window, using the system's WebKitGTK (the .deb installs it). Password, TFA and API-token sign-ins need none of this.