Helix Environment Variables

When deploying Helix with Docker Compose, you can configure various settings using environment variables defined in the .env file. This page describes the key environment variables used to configure the Helix control plane and runner.

Control Plane Environment Variables

The main environment variables for configuring the Helix control plane can be found in the .env.example-prod file:

• KEYCLOAK_ADMIN_PASSWORD: Set the admin password for Keycloak authentication
• POSTGRES_ADMIN_PASSWORD: Set the admin password for the Postgres database
• RUNNER_TOKEN: Set an authentication token for the Helix runner
• KEYCLOAK_FRONTEND_URL: Set the URL where Keycloak authentication is hosted, e.g. http://yourdomain.com/auth/
• SERVER_URL: Set the URL where the Helix server is hosted
• TOGETHER_API_KEY: Set the API key for Together.ai, currently needed for QA pair generation

Additional optional integrations

• GOOGLE_ANALYTICS_FRONTEND: Google Analytics ID
• SENTRY_DSN_FRONTEND / SENTRY_DSN_API: Sentry DSNs for frontend and API monitoring
• EMAIL_SMTP_HOST / EMAIL_SMTP_PORT / EMAIL_SMTP_USERNAME / EMAIL_SMTP_PASSWORD: SMTP config for sending email notifications
• EMAIL_MAILGUN_DOMAIN / EMAIL_MAILGUN_API_KEY: Mailgun config for sending email notifications
• SUBSCRIPTION_QUOTAS_FINETUNING_FREE_MAX_CHUNKS: Set the maximum number of fine-tuning chunks for free users

Runner Environment Variables

Key environment variables for the Helix runner are defined in runner_config.go:

• RUNTIME_AXOLOTL_ENABLED: Enable/disable fine-tuning and inference for Mistral-7B and SDXL models (default true)
• RUNTIME_AXOLOTL_WARMUP_MODELS: Comma-separated list of Mistral-7B and SDXL models to pre-warm, e.g. mistralai/Mistral-7B-Instruct-v0.1,stabilityai/stable-diffusion-xl-base-1.0
• RUNTIME_OLLAMA_ENABLED: Enable/disable LLM inference with Ollama backend (default true)
• RUNTIME_OLLAMA_WARMUP_MODELS: Comma-separated list of LLM models to enable and pre-download for inference, e.g. llama3:instruct,mixtral:instruct
• HF_TOKEN: for text fine tuning which uses Mistral-7B model weights, you now need to accept sharing your contact information with Mistral here and then fetch an access token from here and then specify it in this environment variable

Important notes

• If using Helix on Kubernetes, the Helm chart values like runner.models map to the RUNTIME_OLLAMA_WARMUP_MODELS env var.

The full list of available environment variables can be found for the Control Plane config.go and for the Runner runner_config.go.