> ## Documentation Index
> Fetch the complete documentation index at: https://docs.maition.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Environment Variables (Backend)

> Reference for backend environment variables in .env.rag.example used by the mAItion API and workers.

Reference value for `.env.rag` file.

## Application Setup

* `REDIS_URL`: Redis connection URL
* `POSTGRES_USER`: Postgres username
* `POSTGRES_PASSWORD`: Postgres password
* `POSTGRES_DB`: backend database name
* `POSTGRES_HOST`: Postgres host/service name
* `POSTGRES_PORT`: Postgres port

## LLM Provider

* `OPENROUTER_API_KEY`: API key for OpenRouter/OpenAI-compatible backend
* `OPENROUTER_API_BASE`: base URL for provider API

## Workers and Limits

* `CELERY_CONCURRENCY`: number of Celery worker processes
* `MAX_TASK_CHILD`: max tasks per child process before recycle
* `MAX_MEMORY_PER_CHILD`: per-child memory cap (KB)

## ONNX Runtime / GPU

* `CUDA_VISIBLE_DEVICES`: visible GPU device list
* `ORT_DISABLE_GPU`: disable GPU for ONNX Runtime (`1`/`0`)
* `ORT_DYLD_DISABLE_GPU`: disable GPU for ONNX Runtime on DYLD-based systems

## CORS

* `CORS_ORIGINS`: allowed origins list for cross-origin requests

> NOTE: the value can be provided in a Python-syntax string, e.g. `["http://localhost","http://localhost:3000"]`

## Rate Limiting

* `ENABLE_RATE_LIMIT`: enable rate limiting (`True`/`False`)
* `CHUNK_RATE_LIMIT`: rate limit for chunk endpoint (example: `30/minute`)
* `REPHRASE_RATE_LIMIT`: rate limit for rephrase endpoint (example: `30/minute`)

## MCP Server

* `MCP_ENABLE`: enable the MCP server (`1`/`0`, default: `0`)
* `MCP_API_KEY`: Bearer token clients must present to authenticate. Required when `MCP_ENABLE=1` — the service will not start if this is empty. Generate with `openssl rand -hex 32`