ENVIX

Developer Platform

The open runtime
for serious compute.

Envix Runtime is a CUDA-compatible stack that runs your existing models and pipelines without rewrites. High performance, clear observability, and zero lock-in.

Migration GuideView Documentation

Envix Runtime

Open, CUDA-compatible, production-ready.

Envix Runtime unifies drivers, compiler paths, and orchestration layers into a single coherent stack. It supports all major ML frameworks, runs on standard schedulers, and provides first-class observability out of the box.

CUDA 12.4+ compatible runtime
Open SDK for kernel optimization
Direct device telemetry access
# Check system compatibility
$ envixctl system check
Runtime: 2.4.1
Driver: 560.28.03
CUDA: 12.4
Devices: 8x Envix GPU-80
Status: Ready

# Run validation suite
$ envixctl validate --suite full
[✓] PyTorch 2.4.0
[✓] TensorFlow 2.16.1
[✓] JAX 0.4.30
[✓] vLLM 0.5.3
[✓] Memory bandwidth
[✓] Multi-GPU communication
All checks passed.

Supported Frameworks

Your stack, unchanged.

PyTorch

Full support

Version 2.4+

TensorFlow

Full support

Version 2.16+

JAX

Full support

Version 0.4+

vLLM

Full support

Version 0.5+

ONNX Runtime

Full support

Version 1.18+

Triton

Full support

Version 2.3+

Orchestration

Deploy anywhere.

Docker

Pre-built containers with optimized drivers and runtime

Kubernetes

Native device plugin and GPU operator support

Slurm

First-class scheduler integration with GRES support

Bare Metal

Direct driver installation with minimal dependencies

Developer Tools

Built for operators.

envix-profiler

Kernel-level profiling with memory, compute, and bandwidth analysis

envix-monitor

Real-time GPU metrics, utilization tracking, and alerting

envix-telemetry

OpenTelemetry-compatible observability for production

envix-firmware

Secure firmware updates with rollback support

Migration Guide

Move from NVIDIA in hours.

Envix Runtime is CUDA-compatible. Keep your models, containers, and orchestration config. Swap drivers, validate, and ship.

  1. 1

    Install Envix Runtime

    Single command installs drivers, runtime, and SDK.

  2. 2

    Validate your environment

    Run our validation suite against your existing test cases.

  3. 3

    Deploy to production

    Use your current scheduler. No orchestration changes required.

# Step 1: Install runtime
$ curl -fsSL https://get.envix.com | bash
Installing Envix Runtime 2.4.1...
[✓] Driver 560.28.03
[✓] CUDA Runtime 12.4
[✓] SDK Tools
Installation complete.

# Step 2: Validate
$ envixctl validate --suite torch
Running PyTorch validation...
[✓] Import successful
[✓] CUDA available: True
[✓] Device count: 8
[✓] Memory allocation
[✓] Tensor operations
[✓] Distributed training
All 47 tests passed.

# Step 3: Deploy
$ envixctl deploy --target k8s \
    --namespace ai-prod \
    --replicas 4
Deploying to kubernetes...
[✓] Device plugin ready
[✓] Pods scheduled
[✓] Health checks passing
Deployment complete.

Security Model

Enterprise-grade security.

Secure boot, signed firmware, and least-privilege driver access. Runtime isolation integrates with container security policies and zero-trust network boundaries.

Secure Boot

Cryptographically verified firmware chain

Signed Firmware

All updates cryptographically signed

Runtime Isolation

Container-native security boundaries

Audit Logging

Full device access audit trail

Roadmap Preview

What's coming next.

Q2 2026

  • • Multi-tenant GPU partitioning
  • • Expanded compiler auto-tuning

Q3 2026

  • • PCIe Gen6 support
  • • Distributed training optimizations

Q4 2026

  • • Open telemetry exporters
  • • Extended edge runtime support

Join the Envix Developer Program.

Get early access to runtime updates, direct engineering support, and a community of builders pushing compute forward.

Join Developer Program