Strata

Finds the Strata models on your machine, loads or stops one with a click, and shows the running server's live speed and load.

The Strata screen lists every model it finds, grouped by quant provider (and by model family when there are several), smallest first. Pick one and press Load; Abort cancels a load and Stop ends a running model. Loading one model stops the other first, because each needs the whole GPU; if a Strata server the plugin did not start is running, Load asks you to stop it first. Under the picker, three of Strata's own tools work on the selected model: Benchmark measures the running model's speed, Autotune spends about 20 minutes finding the fastest settings for your PC and saves them for every start (after Cancel it tells you whether anything was saved), and Check for updates installs a new Strata engine from its signed feed (Undo update goes back). Autotune of a running model measures its speed before and after, so you see what it changed. Saved tuning lists the presets a model starts with, with switch off, undo and a 2-minute re-check; Show server log shows its last log lines; and an optional idle stop frees the GPU after a set time without requests. A sidebar row always shows prefill and decode tokens per second, and the screen adds a speed graph, queue, GPU and memory use, and recent requests.

Models are found where Strata itself keeps them: installs named in Strata's settings file or set with STRATA_INSTALL_DIR, and the desktop app's install. On Linux it also finds models run by your own systemd user services, and starts and stops those through their service. Folders with other Strata run configs, and servers on other machines, can be added under Settings, Plugins, Strata, Model discovery, which also lists every place searched. Models outside a systemd service are started with Strata's own command-line tool, through saved profiles named paseo-<model>.

Live status comes from Strata's /metrics, read only from servers whose /health identifies them as Strata. The plugin finds the server by looking for local ports held by a Strata server process, the ports in the found model configs, Strata's saved monitor connections, and ports 8095 and 8080. It polls while a Paseo window shows the sidebar row or the Strata screen.

The plugin reads local files and talks only to Strata servers, on this machine or at addresses you add. It sends nothing else anywhere and has no telemetry. Works on Linux, Windows and macOS with a Strata install; systemd service control is Linux only.