Production Runtime

Stateful RL

Execution Layer

The production runtime for reinforcement learning agents

Fournex is a high-performance execution layer for stateful RL policies inside real systems - robotics, simulations, games, and autonomous services.

In supervised ML, deployment is stateless prediction. In RL, production means long-lived control loops, deterministic execution, and tight environment steps. Fournex provides the runtime for those loops - not a generic model hosting layer.

Talk to the team Read the docs GitHub

Stateful Agent Execution

Run policies across long-lived sessions with persistent state.

Environment Loop Control

Deterministic step and reset cycles with backpressure.

Deterministic Runs

Reproducible execution paths for debugging and safety.

Production Observability

Structured telemetry for policy behavior and system health.

Runtime API

Session Control

Start Session

curl -X POST https://api.fournex.com/sessions -d '{"policy":"agent_abc123"}'

Step Session

curl -X POST https://api.fournex.com/sessions/sess_123/step -d '{"observation":[0.1]}'

# Start an agent session
curl -X POST https://api.fournex.com/sessions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "policy": "agent_abc123",
    "environment": "CartPole-v1",
    "seed": 42
  }'

# Response
{
  "session_id": "sess_123",
  "status": "running",
  "tick": 0
}

Runtime docs Integration guide

Runtime Stack

C + Python

low-latency execution path

Execution Model

Stateful

session-based agent lifecycles

Control Loop

Step / Reset

deterministic environment cycles

Execution Runtime

Stateful Control

Production Systems

Built for real agent systems

Fournex is the runtime layer where RL policies actually run. It handles state, control loops, and production observability so your agents can live inside real systems, not notebooks.

Why RL breaks in production

Deploying agents is not the same as serving predictions.

Long-lived stateful sessions, not stateless calls
Deterministic control loops with timing constraints
Environment step/reset management at scale
Deep integration with robotics, sims, and services

What Fournex is

A production execution layer for reinforcement learning agents.

Stateful policy execution across sessions
Environment loop control with backpressure
Deterministic runs for debugging and safety
Production telemetry for agent behavior

Core capabilities

Designed for high-performance agent execution.

Stateful agent execution
Environment loop management
Session and lifecycle control
Deterministic runs and replay
High-performance inference
Production observability

Who it is for

Teams building real-time agent systems.

Robotics and autonomy engineers
Simulation and game AI teams
Applied reinforcement learning researchers
Infrastructure-minded ML engineers

Not generic ML serving

Built for control loops, not dashboards.

Not a stateless prediction API
Not AutoML or model hosting
Not a visualization or analytics suite
Focused on execution, timing, and state

Example use cases

Where stateful policies need production rigor.

Robotics control loops on real hardware
High-fidelity simulation rollouts
Multi-agent game AI execution
Autonomous agent services with SLAs

Technical positioning

Short, accurate framing for engineers.

A production runtime for RL policies
Where RL policies actually run
An execution layer, not a notebook
From trained model to living system

Fournex is the production runtime for reinforcement learning agents. It provides the execution layer for stateful policies inside real systems - deterministic loops, session control, and observability included.

Talk to the team Read the docs