Free · Self-Hosted · Unlimited Nodes

Install Mesh Optimizer

One command. All platforms. From zero to optimizing in under 5 minutes.

Recommended

One-Line Install

Detects your distro, installs Python if needed, sets up everything, and optionally creates a systemd service.

curl -fsSL https://mesh.slentosystems.com/install.sh | bash

Supports Ubuntu, Debian, Fedora, CentOS, RHEL, Arch, openSUSE, and Pop!_OS. Review the install script source before running.

Advanced Options

curl -fsSL https://mesh.slentosystems.com/install.sh | bash -s -- --license MESH-XXXX-XXXX-XXXX-XXXX

Pre-configure your license key during install. Add --silent for unattended mode, --controller URL to set the controller address.

pip Install

If you already have Python 3.10+ set up:

pip install mesh-optimizer && mesh-optimizer start

Uninstall

curl -fsSL https://mesh.slentosystems.com/installers/linux/uninstall.sh | bash
Recommended

One-Line Install

Works on Intel and Apple Silicon Macs. Installs Homebrew and Python if needed, creates a launchd service for auto-start.

curl -fsSL https://mesh.slentosystems.com/installers/macos/install.sh | bash

Requires macOS 12 (Monterey) or later. Review the install script source before running.

Homebrew

If you prefer Homebrew:

brew install python@3.12 && pip3 install mesh-optimizer

Advanced Options

curl -fsSL https://mesh.slentosystems.com/installers/macos/install.sh | bash -s -- --license MESH-XXXX --silent

Uninstall

curl -fsSL https://mesh.slentosystems.com/installers/macos/uninstall.sh | bash
Recommended

PowerShell Installer

Open PowerShell as Administrator and run this command. Installs Python if needed, sets up everything, and optionally creates a Windows Service.

irm https://mesh.slentosystems.com/installers/windows/install.ps1 | iex

Requires Windows 10/11 or Windows Server 2019+. Review the install script source before running.

Manual Install

If you already have Python 3.10+ installed:

pip install mesh-optimizer
mesh-optimizer start --config mesh_config.yaml

Silent Install

powershell -ExecutionPolicy Bypass -Command "irm https://mesh.slentosystems.com/installers/windows/install.ps1 -OutFile install.ps1; .\install.ps1 -Silent -LicenseKey 'MESH-XXXX'"

Uninstall

irm https://mesh.slentosystems.com/installers/windows/uninstall.ps1 | iex

After Installation

1

Get your free license key

Sign up at portal.slentosystems.com — no credit card required. A community license key is created automatically.

2

Add your license key

Edit ~/.mesh-optimizer/mesh_config.yaml and set your license_key.

3

Choose your JEPA mode ?The JEPA (Joint Embedding Predictive Architecture) is a lightweight ML model that learns your hardware's performance characteristics and makes optimization decisions.

Centralized — The controller trains one global model on data from ALL nodes. Best for clusters where cross-node job routing matters.

Local — Each node trains its own model on local data. Great for edge nodes or when you want fast, independent optimization with no controller latency. ~1% CPU, ~200MB RAM.

Off — No ML optimization. Probes still run and data is still collected, but no ML inference happens on this node. For low-power devices.

Set jepa_mode in your config: "centralized" (default), "local", or "off". Learn more

4

Community atlas sharing ?When enabled, anonymized performance data (kernel throughput, optimal block sizes, memory bandwidth) is shared with the Slento community hub. This data is aggregated to build a global optimization model that benefits all users.

What IS shared: Hardware class (e.g., “RDNA3_24GB”), kernel performance numbers, optimal parameters.

What is NOT shared: Hostnames, IPs, file paths, job commands, or anything identifying.

Set share_atlas_data: false in your config to opt out.

Enabled by default. Your anonymized probe results help improve optimization for everyone. Set share_atlas_data: false to opt out. Learn more

5

Start the agent

Run mesh-optimizer start. The agent detects your hardware, connects to the controller, and begins optimization.

Professional/Enterprise licenses: The controller is automatically downloaded and started — no extra setup needed.

6

Open the dashboard

Navigate to http://localhost:8401 to see your nodes, hardware probes, and optimization status.

Supported Hardware

GPUs

  • AMD Radeon RX 7000 series (RDNA3)
  • AMD Instinct MI200/MI300 (CDNA)
  • NVIDIA GeForce RTX 30/40/50 series
  • NVIDIA Quadro / Tesla / A100

CPUs

  • AMD Ryzen / Threadripper / EPYC
  • Intel Core / Xeon (10th gen+)
  • Apple M1/M2/M3/M4 (via macOS agent)

Other

  • Xilinx Alveo / UltraScale+ FPGAs
  • DDR4/DDR5 memory subsystems
  • NVMe / SSD storage tuning

Requirements

  • Python 3.10+ (auto-installed)
  • 2 GB RAM minimum
  • Network access to controller
  • GPU drivers for GPU optimization

Need Help?

Check the documentation for detailed setup guides, troubleshooting, and API reference.

Documentation Contact Support