[Quixyl-eng]

Summary

• Adds ExVRAM Control, a Tauri window and tray that starts and stops an existing llama-server on 127.0.0.1.
• Memory, context, and KV choices come from the published occupied-8k IQ2_XXS recipe. Unmeasured speed is shown as «Нет данных».
• Protect My PC is on by default: 4 threads, a VRAM leftover, and no killing of other programs. Context compression methods that are not implemented stay off.

Test plan

• cargo test --workspace in control/ (27 core tests)
• existing Python unittest (89 tests, 1 GPU smoke skipped)
• debug panel stayed open with no model loaded and did not spawn llama-server
• click the tray icon and start the published GGUF on a machine with the model and llama-server present

Made with Cursor