Intelligence · NeuroFabric Dynamis · open source

The inference stack, end to end, in Rust

Dynamis replaces every software layer between the GPU and your application — driver, runtime, kernels, scheduler, serving engine and cluster layer — and serves models behind an OpenAI-compatible API.

δύναμις — power, capacity

What it replaces

One stack instead of five.

Layers of the inference stack and what Dynamis replaces
Layer	Usually	With Dynamis
GPU driver	The vendor’s kernel-mode driver	A Rust kernel-mode driver, open for community review; the vendor driver remains supported
Serving engine	vLLM or Ollama	Dynamis serving engine
Cluster layer	Ray	Dynamis cluster layer
Apple Silicon	oMLX	Dynamis on Apple Silicon
API	Varies by engine	OpenAI-compatible

The Rust driver is experimental and released for community review. Production deployments can run Dynamis on the vendor driver.

OpenAI-compatible

Change one base URL.

Dynamis speaks the OpenAI API, so the clients and tools you already use work unchanged, against AEGIS Models or open-weight models.

from openai import OpenAI
import os

client = OpenAI(
    base_url=os.environ["AEGIS_BASE_URL"],
    api_key=os.environ["AEGIS_API_KEY"],
)

for m in client.models.list():
    print(m.id)

// /v1/chat/completions · /v1/embeddings · /v1/models

Open and closed

What is open, and what is not.

NeuroFabric Dynamis

Open source

The stack itself: driver, runtime, kernels, scheduler, serving engine, cluster layer and API.

AEGIS Model Runtime

Proprietary

A closed loader for encrypted AEGIS Model weights, released only to attested machines.

AEGIS Reflex

Proprietary

The closed optimisation plug-in, on NeuroFabric infrastructure only.

About Reflex

Talk to us about your deployment.

Tell us what you need to build and where it must run. A member of the team replies within two business days.

Contact sales How the platform fits