Bring your own model
Paste a Hugging Face repo. Meshive reads the metadata, estimates the VRAM, assigns the cheapest GPU that fits under your price cap, and gives the model its own OpenAI-compatible endpoint.
One required field, and sensible defaults for the rest
- 01
The repo is the whole form
Name, context length, quantization, parameter count and the VRAM estimate are derived from Hugging Face — and every one of them can be overridden when you know better. A fine-tune with no parameter count falls back to its base model's config.
- 02
GGUF included
A GGUF repo registers like any other LLM. The best single variant is chosen automatically, sharded variants are grouped, a multimodal projector is picked up when present, and context length and architecture come from the GGUF metadata.
- 03
Gated repos, without pasting a token twice
Use a one-off token in the form, or store tokens once in the workspace list and pick one by label. Deleting a saved token later does not disturb models already registered with it.
- 04
A registration, not a packaging job
There is no container to build and no manifest to write — the serving engine already speaks the architectures the open ecosystem publishes in. Register the same repo twice and you get the existing registration back rather than a duplicate.
Required, derived, overridable
| Field | Required | Notes |
|---|---|---|
| Hugging Face repo | Yes | The one mandatory input |
| Model name | No | Defaults to the repo's last segment |
| Min VRAM (GB) | No | Overrides the automatic estimate |
| Context length, quantization, parameters | No | Derived from Hugging Face when available |
| vLLM args (JSON) | No | LLMs only |
| Supported sizes / resolutions | Image / video | The API and the Playground validate against this list |
Two limits to know
- Can I register my own image or video model?
- Not right now. Custom diffusion registration is paused while the generation engine moves to ComfyUI. Official image and video models deploy normally, existing custom deployments keep serving, and to run your own workflow today you can deploy ComfyUI on a GPU pod.
- What if the Hugging Face lookup fails?
- Registration proceeds with what you entered — the lookup is best-effort, not a gate. Modality-required fields are still validated, and a diffusion model with no VRAM information at all is rejected rather than registered into something that could never be placed.
Pick a card and start the pod
Sign up, choose the GPU, and the pod is yours in under two minutes. It bills by the hour and stops when you stop it.
Already a user? Invite friends and earn 15% of their first top-up.