Load your first model
Point TurboLLM at the GGUF folders you already have, or browse and download models from Hugging Face without leaving the app. Every model shows a VRAM-fit verdict before you commit and real, measured tokens/sec once it runs.
Reuse the folders you already have
If you've used LM Studio, Ollama, or just downloaded GGUFs by hand, you don't need to move or re-download anything. Point TurboLLM at any directory of GGUF files and it reuses them in place.
Existing GGUF folders are read where they live. TurboLLM indexes them — it does not copy, convert, or fetch a second copy.
Browse & download from Hugging Face, in-app
Search Hugging Face from inside TurboLLM with a live, sortable list next to a permanent detail pane. Pick the model, read its rendered model card, choose a quant, and download — all without switching to a browser.
Live, sortable list
Search and sort results, with a permanent detail pane so you never lose your place while comparing models.
Rendered model card
Read the model's card in-app before you download, so you know what you're getting.
Pick a quant
Choose the quantization that fits your GPU right from the detail pane.
Resume + verify
Downloads resume if interrupted and are checked with SHA-256 verification.
Import from any URL
Not everything lives on Hugging Face. TurboLLM can import a model from any URL, so a GGUF hosted anywhere can be pulled in the same way.
The VRAM-fit verdict
Before you commit to a download, TurboLLM gives you a quant recommendation for your GPU and a VRAM-fit verdict — it shows whether a given quant will actually fit in your available VRAM.
The fit verdict is the same signal that powers auto-tune. See Auto-tune & fit for how the fit is calculated.
Managing your models
Once models are on disk, the manager keeps them organized and honest about performance.
Primary download folder
Set where new downloads land by default.
Pin favourites
Pin the models you use most to the top of the list.
Delete from disk
Remove a model and reclaim the space directly from the manager.
Measured tokens/sec
Every model shows real throughput measured on your machine.
Speed you can trust
The tokens/sec figure next to each model is measured on your machine from actual generation — live while you chat, and last-session when the model is idle. It is never faked or estimated.
TurboLLM only ever shows numbers it has genuinely measured on your hardware. If a model hasn't generated yet, there's simply no speed to show.
Where to go next
Choosing a model and the right quant for your GPU:
- Models hub — GPU-tiered picks for your hardware.
- Quantization guide — what quant to choose and why.
- Auto-tune & fit — how the VRAM-fit verdict works.