Mesh Optimizer

Distributed hardware optimization for heterogeneous compute clusters.

Mesh Optimizer automatically discovers, benchmarks, and optimizes hardware across your infrastructure. Deploy the agent on any machine — it connects to a central controller that orchestrates optimization across your entire fleet.

Supports AMD GPUs, NVIDIA GPUs, Intel/AMD CPUs, FPGAs, and memory subsystems across Linux and macOS.

Quick Start

Install from PyPI

pip install mesh-optimizer

Or install from source

git clone https://github.com/slentoai/mesh-optimizer.git
cd mesh-optimizer
pip install -e .

Configure

cp mesh_config.example.yaml mesh_config.yaml
# Edit mesh_config.yaml — set your controller_url and optional license key

Run

# Start the agent
mesh-optimizer start --config mesh_config.yaml

# Check status
mesh-optimizer status

# View detected hardware
mesh-optimizer hardware

# Stop the agent
mesh-optimizer stop

How It Works

+-------------------+         HTTPS          +---------------------+
|   Mesh Agent      | --------------------> |   Mesh Controller    |
|   (this package)  | <-------------------- |   (proprietary)      |
|                   |                        |                      |
|  - HW discovery   |    register            |  - Job routing       |
|  - Health metrics  |    heartbeat           |  - Optimization      |
|  - Probe runner    |    probe results       |  - Dashboard         |
|  - Job executor    |    job dispatch        |  - Analytics         |
+-------------------+                        +---------------------+

The agent (this package) runs on each node in your cluster. It:

1. Discovers hardware — CPUs, GPUs (AMD and NVIDIA), FPGAs, memory
2. Registers with the controller — sends hardware inventory and metadata
3. Sends heartbeats — CPU/GPU utilization, memory, temperatures, load
4. Runs benchmark probes — measures baseline hardware performance
5. Executes jobs — runs tasks dispatched by the controller

The controller is the central brain that analyzes data from all agents and makes optimization decisions. It is available as a pre-built binary from portal.slentosystems.com.

NAT-Friendly

Agents behind firewalls or NAT work out of the box. Enable NAT mode in your config:

node:
  nat_mode: true
  nat_poll_interval_s: 15

In NAT mode, the agent only makes outbound connections to the controller — no port forwarding required.

Configuration

See mesh_config.example.yaml for all available options.

Key settings:

Setting	Description
controller_url	URL of your Mesh Optimizer controller
node.agent_port	Port for the agent API (default: 8400)
node.nat_mode	Enable for agents behind NAT/firewalls
node.tags	Tags for job routing (e.g., ["gpu", "amd"])
licensing.license_key	License key from the portal (optional)
security.verify_tls	Verify TLS certificates (default: true)

Licensing

Mesh Optimizer is free to use in community mode with unlimited nodes and basic features including hardware discovery, health monitoring, job routing, and dashboard access.

Professional and Enterprise tiers unlock continuous optimization, auto-tuning, and advanced analytics. Get a license key at portal.slentosystems.com.

Automatic Controller Setup

When you add a Professional or Enterprise license key to your config, the agent automatically:

1. Validates your key against the Slento Systems portal
2. Downloads the controller binary for your platform
3. Starts the controller locally alongside the agent
4. Keeps the controller up to date on future launches

No manual controller installation required — just add your license key and start the agent.

licensing:
  license_key: "MESH-XXXX-XXXX-XXXX-XXXX"  # From portal.slentosystems.com

mesh-optimizer start  # Downloads controller, starts everything

API Endpoints

When running in standard mode (not NAT), the agent exposes:

Endpoint	Method	Description
/health	GET	Agent health and metrics
/hardware	GET	Detected hardware inventory
/probe/run	POST	Trigger a benchmark probe
/jobs/submit	POST	Submit a job for execution
/jobs/{id}	GET	Get job status

Requirements

• Python 3.10+
• Linux or macOS
• For GPU detection: nvidia-smi (NVIDIA) or rocm-smi (AMD)
• For FPGA detection: lspci

Documentation

Full documentation is available at docs.slentosystems.com.

License

Apache License 2.0 — see LICENSE for details.

---

The controller is available as a pre-built binary from portal.slentosystems.com.