Skip to content
MZGrid

Ecosystem

Runs the open-source AI stack you already use.

MZGrid is infrastructure, not a walled garden. Workloads are standard containers and inference endpoints speak the OpenAI API, so the tools below work out of the box.

The projects listed here are independent open-source software that is compatible with MZGrid. Their inclusion does not imply a partnership, endorsement or affiliation.

01

Serving & runtimes

The engines our templates are built on.

vllm-openai

vLLM

High-throughput LLM serving with PagedAttention, continuous batching and FP8/AWQ quantisation.

On MZGrid: Powers every LLM template, including llama-3.1-8b, llama-3.3-70b, qwen2.5-32b and the generic vllm-openai template.

Get started
pytorch-jupyter

PyTorch

The standard framework for training and fine-tuning deep learning models.

On MZGrid: The pytorch-jupyter template ships PyTorch 2.5 with CUDA 12.4 in a token-protected JupyterLab.

Get started
sdxl-comfyui

ComfyUI

Node-based diffusion pipelines for image generation and editing.

On MZGrid: The sdxl-comfyui template preloads SDXL and FLUX.1-schnell and exposes the ComfyUI interface and API.

Get started
whisper-large-v3

Whisper

Open speech recognition and translation models.

On MZGrid: The whisper-large-v3 template serves a transcription API you can call from any HTTP client.

Get started

02

Models & packaging

Bring weights and containers from where they already live.

Hugging Face models

The largest open model hub — Llama, Qwen, Mistral, Gemma, embedding models and more.

On MZGrid: Pass any compatible model id to the vllm-openai template; weights are pulled on the provider at start-up.

Get started
custom-docker

Docker

The container format every MZGrid workload runs in.

On MZGrid: Use custom-docker for long-running services or batch for jobs that run to completion. Images run without host networking or privileged mode.

Get started

03

Application frameworks

Your app code doesn’t need to know it’s talking to MZGrid.

OpenAI-compatible clients

The official OpenAI SDKs for Python and TypeScript, and any tool that speaks the same API.

On MZGrid: Set base_url to your job’s endpointUrl + /v1. Chat completions, completions, embeddings and streaming work as-is.

Get started

LangChain

A framework for building LLM-powered applications and agents.

On MZGrid: Use the ChatOpenAI integration with base_url pointed at your MZGrid endpoint.

Get started

LlamaIndex

A data framework for retrieval-augmented generation over your documents.

On MZGrid: Configure the OpenAILike LLM and embedding classes with your MZGrid endpoint URL.

Get started

Ecosystem & developer grants

Building something on MZGrid?

15% of the MZG supply is set aside for ecosystem and developer grants, released linearly over 36 months. Tell us what you’re building in the #builders channel on Discord — integrations, templates, tooling and research that make the network more useful are a priority.