GpuTerm 1.0.8 Beta

Highlights

• Windows remote telemetry: connecting to a Windows host (OpenSSH Server) now populates the full telemetry bar. GpuTerm detects the remote OS per connection (with a cmd.exe /c ver fallback when uname is absent; MSYS/Cygwin count as Windows) and switches to Windows-native sources:
  • CPU: model/core counts from Win32_Processor and a true two-poll usage delta from raw Win32_PerfRawData_PerfOS_Processor counters — locale-independent, unlike Get-Counter paths
  • Memory: Win32_OperatingSystem with page-file usage as swap
  • Disk: Win32_LogicalDisk fixed drives (C:, D:, …)
  • Users: quser, parsed positionally so localized headers don't break it
  • NVIDIA GPUs: full metrics (utilization, VRAM, power, temperature, processes) via nvidia-smi, same as Linux
  • AMD/Intel GPUs: utilization and dedicated VRAM from WDDM GPU performance counters
• Hybrid iGPU+dGPU hosts show both cards — counter data is attributed by adapter LUID from the DirectX registry, and the GPU detail popover appends the integrated GPU alongside the NVIDIA cards. VRAM totals for counter-based GPUs come from the registry's DedicatedVideoMemory.
• All per-tick commands are batched into a single PowerShell 5.1 script sent as -EncodedCommand, so telemetry works whether the OpenSSH default shell is cmd.exe or PowerShell — nothing is installed on the server and no admin rights are required.
• CPU/RAM detail popovers work on Windows too (per-core usage, top processes from a two-sample Get-Process delta, uptime). The detail path now caches the detected remote OS per session instead of re-probing every popover tick.
• Refreshed app icon set (all platforms).

Fixes

• The main window now closes properly from the title-bar close button on every platform. Registering the close-requested handler (which takes detached detail windows along) had disabled Tauri's automatic close, and the follow-up destroy() call lacked the core:window:allow-destroy capability, so the request was silently rejected.

Validation

• 54 frontend tests passed.
• 83 Rust tests passed (33 new Windows parser/detection fixtures).
• cargo clippy is clean; verified against a live Windows host over SSH.

Notes

• Windows remotes need Windows PowerShell 5.1+ (preinstalled on Windows 10/11).
• Load averages don't exist on Windows and show as n/a.
• AMD/Intel GPU power/temperature aren't exposed by the counters (needs Windows 10 1709+ with a WDDM 2.x driver for utilization/VRAM).
• Process owners and GPU process command lines require elevation and fall back to n/a / process names; quser is missing on Home editions, so the Users section stays empty there.
• This is a beta prerelease intended for testing and feedback.

GpuTerm 1.0.7 Beta

Highlights

• macOS (Apple Silicon) telemetry: connecting to a Mac now populates the full telemetry bar. GpuTerm detects the remote OS per connection and switches to macOS-native sources:
  • CPU: chip name with the P/E core layout (e.g. "Apple M2 Pro (8P+4E)"), a true 1-second usage sample from top, and load averages via sysctl
  • Memory: vm_stat + hw.memsize using Activity Monitor's "Memory Used" semantics, plus swap from vm.swapusage
  • Disk: BSD df joined with mount for filesystem types; devfs/autofs are hidden by default
  • Apple GPU: utilization and in-use memory read from ioreg's IOAccelerator statistics — no root required; cards carry an APPLE vendor tag
• CPU/RAM detail popovers work on macOS too (BSD ps top processes, uptime from kern.boottime).
• Remote commands now fall back gracefully on hosts without the timeout binary (macOS 12 and older).

Validation

• 54 frontend tests passed.
• 51 Rust tests passed (13 new macOS parser fixtures).
• TypeScript and Vite production builds passed; cargo clippy is clean.

Notes

• Apple GPU power, temperature, and per-core CPU usage require root powermetrics and are shown as n/a.
• AMD support currently targets rocm-smi.
• This is a beta prerelease intended for testing and feedback.

GpuTerm 1.0.6 Beta

Highlights

• Per-key-type host fingerprints: known_hosts.json now records one SHA-256 fingerprint per host key algorithm, like OpenSSH. A client or server switching algorithms (e.g. RSA → ECDSA) now shows a trust prompt with the key type instead of a false "host key mismatch"; matching legacy entries migrate automatically.
• AMD and Intel GPU telemetry: the bottom bar now monitors AMD GPUs via rocm-smi, Intel discrete GPUs via xpu-smi, and Intel integrated graphics via intel_gpu_top (root or CAP_PERFMON required), alongside NVIDIA. Tools are auto-detected per host, every card shows a vendor tag, and multiple vendors on one machine are collected together.
• ProxyJump: pick another saved profile as a jump host to reach servers behind a bastion. Supports key/agent or password auth for the jump host, verifies each hop's host key, guards against chain loops, and tunnels the terminal, SFTP, telemetry, and detail views alike.
• The SSH library now builds against OpenSSL on Windows, adding curve25519-sha256 and ecdh-sha2-nistp* key exchange so current OpenSSH servers (e.g. Ubuntu 26.04) connect again.
• Added a main-window screenshot to the READMEs.

Validation

• 53 frontend tests passed.
• 37 Rust tests passed.
• TypeScript and Vite production builds passed; cargo clippy is clean.

Notes

• AMD support currently targets rocm-smi; amd-smi-only hosts are detected but not yet parsed.
• Building from source on Windows now requires Perl (Strawberry Perl) for the vendored OpenSSL compile.
• Remote system telemetry remains Linux-first.
• This is a beta prerelease intended for testing and feedback.

GpuTerm 1.0.5 Beta

Highlights

• Multiple concurrent SSH sessions: connecting to a new server no longer drops the current one. Each session keeps its own terminal instance, scrollback, and SFTP path; connected profiles show a green dot in the sidebar and clicking one switches the terminal, SFTP browser, and telemetry to that session.
• Detachable detail windows: every telemetry popover (CPU, Memory, GPU, Disks, Users) has a pop-out button that opens it in a separate OS window. Detached windows refresh independently, close automatically when their session disconnects, and are cleaned up when the main window closes.
• Centered alerts: errors now appear as a centered modal dialog that must be acknowledged, while success and info messages show as a top-center toast that dismisses itself after three seconds.
• Telemetry payloads are tagged with their session id, and background sessions keep their metrics and transfer progress up to date while another session is in view.
• Disconnect now targets the viewed session and falls back to the most recently connected one.
• Updated the English and Korean READMEs for the new capabilities.

Validation

• 49 frontend tests passed.
• 19 Rust tests passed.
• TypeScript and Vite production builds passed; cargo clippy is clean.

Notes

• Remote system telemetry remains Linux-first.
• NVIDIA detail monitoring requires nvidia-smi on the remote host.
• Directory drag-and-drop and transfer resume are not implemented yet.
• This is a beta prerelease intended for testing and feedback.

GpuTerm 1.0.4 Beta

Highlights

• Connection banners (MOTD) are now shown reliably: early terminal output is buffered and replayed after the session attaches, on both first connect and reconnect.
• Multiple session profiles can now be saved: a New button starts a fresh profile and the form shows whether it is editing a saved profile or creating a new one.
• Added a logged-in Users section to the telemetry bar (right of the GPUs) with a detail popover listing user, TTY, login time, and origin.
• The terminal/SFTP split is now adjustable with a drag handle; the chosen width persists across launches and re-clamps automatically on small windows.
• Resource detail popovers (CPU, RAM, Disk, GPU, Users) can be dragged by their title bar and resized from the corner; they hold their position through scrolling, window resizes, and telemetry refreshes.
• The startup warning about a remembered local SFTP path that no longer exists is silenced; explicit browse errors are still reported.
• Rewrote the English and Korean READMEs with a quick-start guide, architecture overview, and troubleshooting reference.

Validation

• 49 frontend tests passed.
• 19 Rust tests passed.
• TypeScript and Vite production builds passed; cargo clippy is clean.

Notes

• Remote system telemetry remains Linux-first.
• NVIDIA detail monitoring requires nvidia-smi on the remote host.
• Directory drag-and-drop and transfer resume are not implemented yet.
• This is a beta prerelease intended for testing and feedback.

GpuTerm 1.0.3 Beta

Highlights

• Added an explicit trust-on-first-use prompt that displays the remote SSH host key fingerprint before saving it.
• Added production and development content security policies for the Tauri window.
• Improved SSH connection timeouts, cleanup, disconnect handling, and blocking task isolation.
• Fixed terminal output when multibyte UTF-8 characters are split across SSH read chunks.
• Stabilized terminal reset, focus, and resize behavior across reconnects and session changes.
• Added working per-file SFTP transfer cancellation.
• Downloads now use temporary files so failed or canceled transfers do not leave partial destination files.
• Improved SFTP error isolation and cleanup during network or session failures.
• Consolidated telemetry parsers and frontend metric formatting into shared, tested utilities.
• Expanded telemetry, GPU selector, disk priority, SFTP, formatting, and terminal stream tests.

Validation

• 38 frontend tests passed.
• 17 Rust tests passed.
• TypeScript and Vite production build passed.

Notes

• Remote system telemetry remains Linux-first.
• NVIDIA detail monitoring requires nvidia-smi on the remote host.
• SFTP transfers can be canceled but cannot yet be resumed.
• This is a beta prerelease intended for testing and feedback.

GpuTerm 1.0.2 Beta

Highlights

• Added clickable CPU, RAM, GPU, and Disk telemetry details.
• Added CPU per-core usage, load, uptime, and top-process views.
• Added detailed RAM, swap, cache, buffer, and top-process views.
• Added NVIDIA GPU detail tabs with utilization, VRAM, power, temperature, clocks, and compute processes.
• Fixed multi-GPU selection by using controlled UUID-based state.
• GPU selection now survives telemetry refreshes and detail popover reopening, with safe fallback when a GPU disappears.
• GPU selection resets cleanly when switching SSH sessions.
• Kept telemetry collection isolated from the interactive SSH terminal channel.

Validation

• 25 frontend tests passed.
• 11 Rust tests passed.
• TypeScript and Vite production build passed.

Notes

• Remote system telemetry is Linux-first.
• Detailed GPU metrics require NVIDIA drivers and nvidia-smi on the remote host.
• This is a beta prerelease intended for testing and feedback.

GpuTerm 1.0.1 Beta

GpuTerm 1.0.1 Beta

Beta release focused on disk telemetry details and SFTP transfer UX.

Highlights

• Added a clickable Disk summary in the telemetry bar that opens a full disk detail popover.
• Added disk detail fields for mount point, filesystem, filesystem type, total, used, available, and usage percentage.
• Added disk usage progress bars with warning state at 80%+ and critical state at 90%+.
• Added a "Show hidden filesystems" toggle in the disk detail popover.
• Fixed the disk detail popover so it renders outside the telemetry bar and is not clipped by footer overflow.
• Added SFTP drag-and-drop upload from the local file panel to the remote panel.
• Added SFTP drag-and-drop download from the remote file panel to the local panel.
• Added a TransferQueue UI with filename, direction, source path, target path, progress, status, and per-file errors.
• Added overwrite confirmation before upload/download when the target file already exists.
• Changed SFTP transfer streaming to use 1 MiB chunks.
• Added backend command placeholders for transfer cancellation.
• Updated app/package versions to 1.0.1-beta.

Notes

• Directory drag-and-drop is detected but not supported yet.
• Transfer cancellation UI and command shape are present, but active stream cancellation is not wired in this beta.
• After updating from source, fully restart npm run tauri:dev so Rust commands and Tauri configuration changes are reloaded.

Verification

• npm run test: 16 tests passed.
• npm run build: passed with the existing Vite chunk-size warning.
• cargo test: 7 tests passed.