Live GPU telemetry, a real SSH terminal, Docker control, chat with your own local LLM and one-tap model downloads — everything your Spark does, on the phone in your hand.
iPhone & iPad · Android coming · 9 languages · connects over SSH on your own network












What it does
Machete talks to your machine over SSH on your own network. No cloud in the middle, no agent required to get started.
Utilisation, temperature, power draw, unified memory, disk and network — sampled continuously, with history charts and a clear marker when data goes stale.
Every process resident in GPU memory, with its footprint and endpoint — and one tap to unload a model and free the memory.
A proper xterm session with a key toolbar, not a command box. Paste, scroll, Ctrl-combos.
Containers and images with live CPU, memory and per-container GPU usage. Start, stop, restart, inspect logs.
Any OpenAI-compatible endpoint on your Spark — vLLM, llama.cpp. Streamed replies, model picker, throughput benchmark.
Search the hub, download straight to the Spark with resumable multi-connection transfers, then serve it with a saved run config.
Finds your running fine-tune, plots the loss curve, estimates time left and what the run has cost in electricity.
Thresholds on any metric, plus training finished, training stalled and loss targets — delivered as notifications.
Configs, Docker volumes and any path you choose, on a schedule, to the cloud storage you already use.
Machete Pro
The live dashboard, GPU telemetry and history stay free forever. A Pro subscription unlocks everything that reaches into the machine.
Monthly or yearly, billed through the App Store. Cancel any time in your account settings — cancellation takes effect at the end of the current period.