Slento Systems Documentation

Welcome to the Slento Systems documentation. Here you'll find everything you need to set up, configure, and manage your infrastructure optimization mesh and pharma discovery platform.

Products

Quick Start

Get a mesh agent running in 3 steps:

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

# ~/.mesh-optimizer/config.yaml
license_key: MESH-XXXX-XXXX-XXXX-XXXX
controller_url: http://your-controller:8401
node_name: my-workstation

mesh-optimizer start

Installation

System Requirements

Automatic Installation

The recommended way to install on Linux and macOS:

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

This will:

• Detect your OS and package manager
• Install Python 3.9+ if not present
• Create a virtual environment at ~/.mesh-optimizer/
• Install all dependencies
• Generate a default config file
• Optionally create a systemd/launchd service

Manual Installation

python3 -m venv ~/.mesh-optimizer/venv
source ~/.mesh-optimizer/venv/bin/activate
pip install mesh-optimizer

GPU Support

Configuration

The agent reads configuration from ~/.mesh-optimizer/config.yaml.

# Node identity
node_name: my-workstation
license_key: MESH-XXXX-XXXX-XXXX-XXXX

# Controller connection
controller_url: http://controller-ip:8401
heartbeat_interval: 10  # seconds

# Networking
nat_mode: false          # Set true if behind NAT
public_url: ""           # Your public URL if using NAT
nat_poll_interval: 5     # seconds between job polls in NAT mode

# Agent API (disabled in NAT mode)
agent_port: 8400

# Probing
probe_interval: 21600    # seconds (6 hours)
probe_types:             # which probes to run
  - bandwidth
  - compute
  - latency
  - memory

# Resources
max_concurrent_jobs: 2
gpu_devices: auto        # "auto", "all", or list of indices

Environment Variables

All config options can also be set via environment variables with the MESH_ prefix:

export MESH_LICENSE_KEY=MESH-XXXX-XXXX-XXXX-XXXX
export MESH_CONTROLLER_URL=http://controller:8401
export MESH_NAT_MODE=true

How Mesh Optimizer Works

The Mesh Optimizer is a distributed system with three components:

Architecture

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│   Agent #1   │     │   Agent #2   │     │   Agent #3   │
│  (7900 XTX)  │     │  (RTX 4090)  │     │  (Xeon CPU)  │
└──────┬───────┘     └──────┬───────┘     └──────┬───────┘
       │ heartbeat          │ heartbeat          │ heartbeat
       │ atlas sync         │ atlas sync         │ atlas sync
       └────────────────────┼────────────────────┘
                            │
                    ┌───────▼───────┐
                    │  Controller   │
                    │  (JEPA Model) │
                    │  (Job Router) │
                    │  (Dashboard)  │
                    └───────────────┘

Data Flow

1. Agents boot, scan hardware, and register with the controller
2. Every 10 seconds: heartbeat with CPU, memory, GPU, and disk metrics
3. Every 6 hours: run optimization probes and sync results to the controller
4. Controller feeds probe data to the JEPA model for online learning
5. When a job is submitted, the JEPA model predicts the best node
6. Job is routed, executed, and results returned

Agent Setup

Each machine in your mesh runs an agent. The agent handles:

• Hardware auto-discovery (CPUs, GPUs, FPGAs, memory)
• Health reporting (CPU/GPU utilization, memory, temperatures)
• Performance probing (bandwidth, compute, latency benchmarks)
• Job execution (runs workloads routed by the controller)

Starting the Agent

# Start in foreground
mesh-optimizer start

# Start as a service (Linux)
sudo systemctl enable mesh-optimizer
sudo systemctl start mesh-optimizer

# Check status
mesh-optimizer status

Hardware Detection

The agent automatically detects:

Controller

The controller is the brain of the mesh. It runs on one machine (typically the most capable) and coordinates all agents.

Starting the Controller

# Start controller (also starts a local agent)
mesh-optimizer controller start

# The controller runs three services:
#   Port 8400 - Local agent API
#   Port 8401 - Controller API (agents connect here)
#   Port 8402 - Dashboard (web UI)

JEPA Model

The controller trains a Joint Embedding Predictive Architecture (JEPA) model that learns hardware performance characteristics across your fleet. It uses this model to:

• Predict which node will perform best for a given workload
• Detect performance anomalies
• Recommend optimal configurations

The model trains automatically from probe data. You can trigger manual retraining from the dashboard.

JEPA Mode

Each node can be configured to run the JEPA optimization engine in one of three modes. Set jepa_mode in your node config:

# In mesh_config.yaml
node:
  jepa_mode: "centralized"   # or "local" or "off"

Failover

If the controller goes down, agents elect a new controller automatically:

• Agents detect missing heartbeat after 2 minutes
• Leader election based on: load, GPU capability, and uptime
• New controller announces to all peers
• Original controller yields when it comes back online

Community Atlas Sharing

Mesh Optimizer can share anonymized performance data with the Slento community hub to improve optimization for all users. This is enabled by default and can be disabled at any time.

How It Works

1. Your agent runs hardware probes and collects performance data (kernel throughput, optimal block sizes, memory bandwidth, etc.)
2. Every 6 hours, the data is anonymized — hostnames, IPs, file paths, and job commands are stripped
3. Only the hardware class (e.g., “RDNA3_24GB”, “Xeon_16C”) and performance metrics are sent to the community hub
4. The hub aggregates data from all contributors to train a global optimization model
5. Your controller periodically pulls model improvements back, enriching your local atlas with insights from hardware you don’t have

What Is Shared

Configuration

# In mesh_config.yaml
node:
  # Set to false to opt out of community sharing
  share_atlas_data: true   # default: enabled

API Endpoints

• GET /community/stats — View your community sync status (last push/pull times, points contributed)
• POST /community/ingest — Receives aggregated community data from the hub (called by the hub, not by users)
• GET /nodes/{node_id}/atlas-export — Export anonymized atlas data for a node

Networking

LAN Mode (Default)

All nodes on the same network. Controller connects directly to agent APIs.

# Agent config
controller_url: http://192.168.1.100:8401
agent_port: 8400

NAT/WAN Mode

For nodes behind firewalls or on different networks. Agents poll the controller for jobs instead of accepting inbound connections.

# Agent config
controller_url: https://mesh.yourcompany.com:8401
nat_mode: true
nat_poll_interval: 5

Hybrid Mode

Mix LAN and WAN nodes in the same mesh. LAN nodes use direct connections; WAN nodes use NAT mode.

Ports

Dashboard

The web dashboard is available at http://controller:8402 and provides:

• Node Grid — Live health status of all nodes with CPU, GPU, memory gauges
• Node Detail — Per-node performance history, hardware inventory, probe results
• Jobs — Job queue, history, execution times, and routing decisions
• Atlas — Federated atlas data across all nodes
• JEPA — Model confidence, training history, and manual retrain

Pharma Discovery

The Pharma Discovery platform uses AI to accelerate drug discovery workflows. It combines multi-property landscape exploration with virtual screening against 15,000+ ChEMBL targets.

Key Capabilities

• SSI Landscape Exploration — Navigate chemical space using learned property landscapes
• Virtual Screening — Screen compound libraries against validated targets
• ADMET Prediction — Predict absorption, distribution, metabolism, excretion, and toxicity
• Molecular Scoring — Multi-property optimization with configurable objectives
• Computational Chemistry — Binding affinity estimation and conformational analysis

Exploration Runs

An exploration run navigates chemical space to find novel compounds matching your target profile. Each run:

1. Defines a target property landscape (e.g., high binding + low toxicity)
2. Generates and evaluates candidate molecules using the SSI swarm
3. Returns a ranked list of promising compounds with predicted properties

Exploration credits are consumed per run. See Tiers & Pricing for credit allocations.

Virtual Screening

Screen your compound library against validated target models. Each screening credit covers one compound evaluation.

Supported target databases:

• ChEMBL (15,000+ targets)
• Custom target models (Enterprise tier)

Pharma API Reference

Check Usage & Credits

GET /api/pharma-usage.php?key=PHRM-XXXX-XXXX-XXXX-XXXX

Response:
{
  "license_key": "PHRM-XXXX-XXXX-XXXX-XXXX",
  "tier": "base",
  "credits": {
    "exploration": {"total": 5, "used": 2, "remaining": 3},
    "screening": {"total": 55000, "used": 12300, "remaining": 42700}
  },
  "usage": {
    "score": {"used": 1200, "limit": 5000},
    "screen": {"used": 0, "limit": 1}
  }
}

Record Usage

POST /api/pharma-usage.php
Content-Type: application/json

{
  "key": "PHRM-XXXX-XXXX-XXXX-XXXX",
  "action": "score",
  "count": 100
}

Response:
{
  "status": "recorded",
  "remaining": {"score": 3800}
}

Tiers & Pricing

Mesh Optimizer

Pharma Discovery

Credit Packs (Add-On)

License Activation

After purchasing, your license key appears in the portal. Add it to your config:

# Mesh Optimizer
license_key: MESH-XXXX-XXXX-XXXX-XXXX

# Pharma Discovery
license_key: PHRM-XXXX-XXXX-XXXX-XXXX

The agent validates the key on startup and periodically during operation.

License Validation API

POST /api/validate.php
Content-Type: application/json

{
  "license_key": "MESH-XXXX-XXXX-XXXX-XXXX",
  "hardware_id": "optional-machine-fingerprint"
}

Response:
{
  "valid": true,
  "tier": "professional",
  "max_nodes": 25,
  "expires_at": "2027-03-08T22:39:01Z",
  "product": "mesh_optimizer"
}

REST API Reference

Agent API (Port 8400)

Controller API (Port 8401)

Troubleshooting

Agent won't connect to controller

• Verify the controller_url is reachable: curl http://controller:8401/health
• Check firewall rules — port 8401 must be open on the controller
• If behind NAT, set nat_mode: true

GPU not detected

• AMD: Ensure rocm-smi or /sys/class/drm/card*/device/vendor is accessible
• NVIDIA: Ensure nvidia-smi is in PATH
• Check that the agent user has permission to access GPU devices

License validation fails

• Verify the key in your config matches what's shown in the portal
• Check that your license hasn't expired
• Ensure the agent can reach portal.slentosystems.com on port 443

High latency on heartbeats

• Increase heartbeat_interval if on a slow connection
• Check network latency with ping controller-ip

Changelog

v1.0.0 — March 2026

• Initial release
• Mesh Optimizer with JEPA-driven job routing
• Pharma Discovery with SSI exploration and virtual screening
• Support for AMD (RDNA/CDNA), NVIDIA, Intel GPUs
• NAT/WAN mode for distributed deployments
• Web dashboard with live monitoring
• Automatic failover with leader election