Skip to content

Configuration Guide

Generated: 2025-12-28 | Author: mbaetiong

Overview

CODEX uses a hierarchical configuration system supporting multiple environments (development, staging, production) with secure secret management.

Configuration Files

Primary Config: config/config.yaml

# Environment-specific settings
environment: ${ENV:-development}

# Model configuration
model:
  registry_uri: ${MLFLOW_TRACKING_URI}
  artifact_location: ${ARTIFACT_STORE}
  default_model: "codex-classifier-v2"

# Feature engineering
features:
  enable_cache: true
  cache_ttl_seconds: 3600

# Inference
inference:
  batch_size: 32
  timeout_seconds: 30
  max_retries: 3

# Monitoring
monitoring:
  enable_prometheus: true
  metrics_port: 9090
  log_level: ${LOG_LEVEL:-INFO}

Environment Variables (.env)

# Core settings
ENV=production
LOG_LEVEL=INFO

# MLflow
MLFLOW_TRACKING_URI=https://mlflow.codex.ai
MLFLOW_EXPERIMENT_NAME=codex-experiments

# AWS credentials (use Secrets Manager in production)
AWS_ACCESS_KEY_ID=<from-secrets-manager>
AWS_SECRET_ACCESS_KEY=<from-secrets-manager>

# Database
DATABASE_URL=postgresql://user:pass@host:5432/codex

Configuration Priority (Highest to Lowest)

  1. Environment Variables: export KEY=value
  2. .env file: Local development overrides
  3. config/config.yaml: Default settings
  4. Secrets Manager: Production secrets (AWS Secrets Manager, GitHub Secrets)

Secrets Management

Development

  • Use .env file (never commit to Git, add to .gitignore)

Production

  • GitHub Actions: Use ${{ secrets.SECRET_NAME }}
  • AWS: Use Secrets Manager with IAM role-based access
  • Kubernetes: Use Sealed Secrets or External Secrets Operator

Example (GitHub Actions): 

- name: Deploy model
  env:
    MLFLOW_TRACKING_URI: ${{ secrets.MLFLOW_URI }}
    AWS_ACCESS_KEY_ID: ${{ secrets.AWS_KEY }}
  run: python deploy.py

Loading Configuration in Code

from codex_ml.config import load_config

# Load configuration (auto-detects environment)
config = load_config()

# Access settings
model_name = config.model.default_model
batch_size = config.inference.batch_size

# Override specific settings
config.override("inference.timeout_seconds", 60)

See Also