HM Studio Runner

Local MLX model runner is up.

This is the native localhost daemon for model loading and generation. You can register a model, load it, and send a prompt directly from this page.

GPU-first runner Checking whether MLX is using GPU or CPU.

MLX Starter Cards

Pick a proven MLX starter to prefill the centered studio below. GPU is the intended path for interactive use.

GGUF Starter Cards

Use these when you want the `llama.cpp` / GGUF path. They register `gguf-llm` manifests and expect `llama-server` to be installed first.

Registered Models

This is the current in-memory registry state from /runner/v1/models and /daemon/status.

Catalog Search

Register remote nodes, search by machine model, lane, memory, location, or model id, and ask the runner for the current best fit.

No catalog action yet.
No catalog results yet.
No best-node result yet.

Audio To Text

Paste the faster-whisper environment path once. The runner assumes .venv/bin/python, tests the config, loads the long-lived ASR worker, then runs transcription against your uploaded file.

Paste faster-whisper environment path to begin.
No ASR runtime logs yet.
No transcript yet.

Load And Configure

Load and unload control in-memory residency. Cache clearing removes downloaded model files from disk.

No action yet.

Certification

Runs a cold download/load, a cold generate, a warm reload, and a second generate using the selected model.

No validation run yet.
No certification report yet.

Chat

Ready.
No response yet.