Quickstart

This guide will get you running your first MOSAIC experiment in minutes.

Prerequisites

Before launching, make sure:

  1. You have completed the Installation guide

  2. Your virtual environment is activated:

    source .venv/bin/activate

  3. The .env file exists in the project root (copy from the example if needed):

    cp .env.example .env

    The .env file controls Qt rendering, GPU settings, environment defaults, API keys, and worker-specific configuration. See the comments inside .env.example for details.

Launching MOSAIC

MOSAIC uses a two-process architecture: a Trainer Daemon (gRPC backend) and the Qt6 GUI (frontend). The launch script handles both:

./run.sh

What run.sh does:

  1. Loads environment variables from .env

  2. Activates the virtual environment (if .venv/ exists)

  3. Creates var/logs/ and var/trainer/ directories

  4. Kills any stale trainer daemon processes

  5. Starts the Trainer Daemon in the background (gym_gui.services.trainer_daemon on 127.0.0.1:50055)

  6. Waits for the daemon to be ready (gRPC health check, up to 10 retries)

  7. Launches the MOSAIC GUI (gym_gui.app)

On exit, the script automatically shuts down the daemon.

Tip

To use a specific Python version:

PYTHON_BIN=python3.11 ./run.sh

Running a Single-Agent Environment

  1. Select the Single Agent tab

  2. Choose an environment family and environment, for example: Gymnasium > Classic Control > CartPole-v1

  3. Click Load Environment

  4. Select Human Keyboard as the actor

  5. Click Play and use arrow keys to control the cart

Running a Multi-Agent Environment

  1. Select the Multi Agent tab

  2. Choose an environment, for example: PettingZoo > Classic > Chess

  3. Click Load Environment

  4. Configure agents:

    • Player 0: Human Keyboard

    • Player 1: Random Policy

  5. Click Play

Project Structure

After launching, MOSAIC creates runtime directories under var/:

var/
├── logs/
│   └── trainer_daemon.log    # Daemon stdout/stderr
├── trainer/
│   └── trainer.pid           # Daemon PID file
└── operators/
    └── telemetry/            # Per-run JSONL telemetry

Logs are the first place to look if something goes wrong.

Troubleshooting

Daemon fails to start

Check var/logs/trainer_daemon.log for errors. Common causes:

  • Port 50055 already in use (another daemon running)

  • Missing .env file or incorrect QT_API setting

  • gRPC/protobuf version mismatch

GUI shows a blank window

On WSL, make sure these are set in .env:

QT_QPA_PLATFORM=xcb
QT_QUICK_BACKEND=software
LIBGL_ALWAYS_SOFTWARE=1
“No module named gym_gui”

Make sure MOSAIC is installed in development mode:

pip install -e .

Next Steps

  • Workers – how workers train agents

  • Operators – how operators evaluate agents