Skip to main content

Environment Variables

Configure your Chiro AI instance behavior through environment variables. Achiral supports both predefined system variables and custom application variables.

Setting Environment Variables​

Via Dashboard​

  1. Navigate to Configuration → Environment Variables
  2. Click Add Variable
  3. Enter the variable name and value
  4. Click Save
  5. Restart your Chiro instance to apply changes

Via API​

curl -X POST https://api.achiral.ai/v1/organizations/{org_id}/env \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "variables": {
      "LOG_LEVEL": "DEBUG",
      "MAX_CONCURRENT_REQUESTS": "100"
    }
  }'

System Variables​

Core Configuration​

MAX_CONCURRENT_REQUESTS​

Maximum number of simultaneous inference requests.

  • Default: Varies by plan tier (5/20/100 for Spark/Seed/Scale)
  • Type: Integer
  • Range: 1-1000
  • Example: MAX_CONCURRENT_REQUESTS=100

REQUEST_TIMEOUT​

Maximum time (in seconds) to process a single request.

  • Default: 120
  • Type: Integer
  • Range: 10-600
  • Example: REQUEST_TIMEOUT=60

MODEL_CACHE_DIR​

Directory for cached model files and weights.

  • Default: /var/lib/chiro/models
  • Type: String (path)
  • Example: MODEL_CACHE_DIR=/mnt/storage/models

WORKER_PROCESSES​

Number of inference worker processes.

  • Default: Auto (based on plan tier)
  • Type: Integer
  • Range: 1-16
  • Example: WORKER_PROCESSES=4

Logging Configuration​

LOG_LEVEL​

Logging verbosity level.

  • Default: INFO
  • Type: String
  • Options: DEBUG, INFO, WARNING, ERROR, CRITICAL
  • Example: LOG_LEVEL=DEBUG

LOG_FORMAT​

Format for log output.

  • Default: json
  • Type: String
  • Options: json, text, structured
  • Example: LOG_FORMAT=json

LOG_DESTINATION​

Where to send logs.

  • Default: stdout
  • Type: String
  • Options: stdout, file, syslog, cloudwatch
  • Example: LOG_DESTINATION=file

LOG_FILE_PATH​

Path to log file (when LOG_DESTINATION=file).

  • Default: /var/log/chiro/app.log
  • Type: String (path)
  • Example: LOG_FILE_PATH=/mnt/logs/chiro.log

Model Configuration​

DEFAULT_MODEL​

Default model to use when not specified in requests.

  • Default: None (must be specified in request)
  • Type: String
  • Example: DEFAULT_MODEL=mistral-7b-instruct

MODEL_AUTO_DOWNLOAD​

Automatically download models from Hugging Face Hub.

  • Default: true
  • Type: Boolean
  • Example: MODEL_AUTO_DOWNLOAD=false

MAX_MODEL_SIZE_GB​

Maximum allowed model size in gigabytes.

  • Default: Varies by plan tier
  • Type: Integer
  • Example: MAX_MODEL_SIZE_GB=50

MODEL_QUANTIZATION​

Default quantization mode for models.

  • Default: none
  • Type: String
  • Options: none, int8, int4, gptq, awq
  • Example: MODEL_QUANTIZATION=int8

API Configuration​

API_PORT​

Port number for the API server.

  • Default: 8000
  • Type: Integer
  • Range: 1024-65535
  • Example: API_PORT=9000

API_HOST​

Host address to bind the API server.

  • Default: 0.0.0.0
  • Type: String (IP address)
  • Example: API_HOST=127.0.0.1

API_KEY_REQUIRED​

Require API key for all requests.

  • Default: true
  • Type: Boolean
  • Example: API_KEY_REQUIRED=true

CORS_ORIGINS​

Allowed CORS origins (comma-separated).

  • Default: *
  • Type: String
  • Example: CORS_ORIGINS=https://app.example.com,https://admin.example.com

RATE_LIMIT_REQUESTS​

Maximum requests per minute per API key.

  • Default: 100
  • Type: Integer
  • Example: RATE_LIMIT_REQUESTS=500

Performance Tuning​

BATCH_SIZE​

Default batch size for inference.

  • Default: Auto (optimized per plan tier)
  • Type: Integer
  • Range: 1-256
  • Example: BATCH_SIZE=32

GPU_MEMORY_FRACTION​

Fraction of GPU memory to allocate.

  • Default: 0.9
  • Type: Float
  • Range: 0.1-1.0
  • Example: GPU_MEMORY_FRACTION=0.8

ENABLE_FLASH_ATTENTION​

Use Flash Attention for faster inference.

  • Default: true
  • Type: Boolean
  • Example: ENABLE_FLASH_ATTENTION=true

KV_CACHE_SIZE_GB​

Size of key-value cache in gigabytes.

  • Default: Auto (based on available VRAM)
  • Type: Integer
  • Example: KV_CACHE_SIZE_GB=8

Training Configuration​

TRAINING_ENABLED​

Enable model training and fine-tuning features.

  • Default: true
  • Type: Boolean
  • Example: TRAINING_ENABLED=false

CHECKPOINT_INTERVAL​

Save training checkpoints every N steps.

  • Default: 1000
  • Type: Integer
  • Example: CHECKPOINT_INTERVAL=500

GRADIENT_ACCUMULATION_STEPS​

Number of steps to accumulate gradients.

  • Default: 4
  • Type: Integer
  • Example: GRADIENT_ACCUMULATION_STEPS=8

Custom Variables​

You can define custom environment variables for your application:

# Application-specific variables
APP_NAME=my-llm-app
CUSTOM_API_ENDPOINT=https://api.example.com
FEATURE_FLAG_STREAMING=true

Access custom variables in your application code:

import os

app_name = os.getenv('APP_NAME', 'default-app')
api_endpoint = os.getenv('CUSTOM_API_ENDPOINT')

Variable Management Best Practices​

Security​

  • Never commit secrets: Use the Achiral secrets management system for sensitive data
  • Rotate regularly: Update API keys and tokens on a regular schedule
  • Limit access: Restrict who can view and modify environment variables

Organization​

  • Use prefixes: Group related variables with prefixes (e.g., DB_, API_, CACHE_)
  • Document purpose: Add descriptions for custom variables in your documentation
  • Version control: Track changes to environment variables in your deployment pipeline

Performance​

  • Cache frequently accessed values: Don't read environment variables in hot paths
  • Validate on startup: Check required variables exist and have valid values at startup
  • Use defaults wisely: Provide sensible defaults for optional configuration

Environment Variable Precedence​

Variables are loaded in the following order (later sources override earlier ones):

  1. System defaults
  2. Plan-specific defaults
  3. Organization-level variables
  4. Chiro instance variables
  5. Runtime overrides (via API)

Viewing Current Variables​

Via Dashboard​

Navigate to Configuration → Environment Variables to view all active variables and their values.

Via API​

curl https://api.achiral.ai/v1/organizations/{org_id}/env \
  -H "Authorization: Bearer YOUR_API_KEY"

Deleting Variables​

Via Dashboard​

  1. Navigate to Configuration → Environment Variables
  2. Find the variable to delete
  3. Click the trash icon
  4. Confirm deletion

Via API​

curl -X DELETE https://api.achiral.ai/v1/organizations/{org_id}/env/VARIABLE_NAME \
  -H "Authorization: Bearer YOUR_API_KEY"

Next Steps​