Configuration Reference

All orchestrator configuration is supplied via environment variables. These can be set in docker-compose.yaml, Kubernetes ConfigMaps, or directly in your shell.

Core Settings

ENVIRONMENT

Type: string
Default: production
Values: production | development Runtime environment label. Affects logging verbosity and data directory defaults.

ENVIRONMENT=development

LOG_LEVEL

Type: string
Default: info
Values: debug | info | warn | error Log verbosity level. Use debug for troubleshooting, info for production.

LOG_LEVEL=debug

PORT

Type: string
Default: 8080 HTTP port for the orchestrator API and MCP endpoints.

PORT=8080

STORAGE_DIR

Type: string
Default: /app/data (or {ENVIRONMENT}-data if set) Directory path for the SQLite database and file storage. Must be writable by the orchestrator process.

STORAGE_DIR=/app/data

In Kubernetes, this should match the PersistentVolumeClaim mount path.

Worker Pool Configuration

These settings control how the orchestrator spawns and manages Kubernetes worker jobs.

MIN_WORKERS

Type: int
Default: 1 Minimum number of concurrent worker jobs. The orchestrator maintains at least this many workers when tasks are queued.

MIN_WORKERS=1

MAX_WORKERS

Type: int
Default: 10 (Docker Compose) / 50 (config default) Maximum number of concurrent worker jobs across all clusters. Prevents resource exhaustion.

MAX_WORKERS=20

QUEUE_THRESHOLD

Type: int
Default: 5 Number of queued tasks before the orchestrator spawns an additional worker. Lower values spawn workers more aggressively.

QUEUE_THRESHOLD=5

WORKER_NAMESPACE

Type: string
Default: mimir-aip (Helm uses release namespace) Kubernetes namespace where worker jobs are created. Must match the orchestrator’s namespace for RBAC permissions.

WORKER_NAMESPACE=mimir-aip

WORKER_SERVICE_ACCOUNT

Type: string
Default: mimir-worker (Helm) / worker-service-account (config) Kubernetes ServiceAccount assigned to worker jobs. Must have permissions to access required resources.

WORKER_SERVICE_ACCOUNT=mimir-worker

WORKER_IMAGE

Type: string
Default: mimir-aip/worker:latest Docker image used for worker jobs. Set this when using a custom registry or version.

WORKER_IMAGE=ghcr.io/mimir-aip/worker:0.1.1

WORKER_CPU_LIMIT

Type: string
Default: 2000m Kubernetes CPU limit for worker pods. Format: millicores (e.g., 2000m = 2 CPU cores).

WORKER_CPU_LIMIT=4000m

WORKER_MEMORY_LIMIT

Type: string
Default: 4Gi Kubernetes memory limit for worker pods. Format: standard units (e.g., 4Gi, 512Mi).

WORKER_MEMORY_LIMIT=8Gi

Multi-Cluster Configuration

CLUSTER_CONFIG_FILE

Type: string (file path)
Default: "" (empty = single in-cluster mode) Path to a YAML file defining multiple Kubernetes clusters for worker dispatch. When unset, the orchestrator uses only the in-cluster configuration.

CLUSTER_CONFIG_FILE=/etc/mimir/clusters.yaml

Example clusters.yaml:

clusters:
  - name: primary
    kubeconfig: ""
    namespace: mimir-aip
    orchestratorURL: http://mimir-aip-orchestrator:8080
    maxWorkers: 50
    serviceAccount: mimir-worker

  - name: site-b
    kubeconfig: /etc/mimir/kubeconfigs/site-b.yaml
    namespace: mimir-aip
    orchestratorURL: http://192.168.10.5:8080
    maxWorkers: 100
    serviceAccount: mimir-worker

WORKER_AUTH_TOKEN

Type: string (Bearer token)
Default: "" (empty = authentication disabled) Shared secret token for worker-to-orchestrator authentication. Workers include this as Authorization: Bearer <token> when calling /api/worktasks/* endpoints.

WORKER_AUTH_TOKEN=your-secure-random-token-here

Keep this token secret. Store it in Kubernetes Secrets or a secure vault.

ORCHESTRATOR_URL

Type: string (URL)
Default: http://localhost:8080 URL where workers can reach the orchestrator API. In Kubernetes, this is typically the orchestrator Service name.

ORCHESTRATOR_URL=http://mimir-aip-orchestrator:8080

Advanced Worker Settings

WORKER_CONCURRENCY_LIMITS

Type: string (JSON object)
Default: See below Per-task-type concurrency limits. Prevents a single task type from consuming all workers.

WORKER_CONCURRENCY_LIMITS='{"ml_training":5,"ml_inference":10,"pipeline_execution":20,"digital_twin_update":10}'

Default limits:

{
  "ml_training": 5,
  "ml_inference": 10,
  "pipeline_execution": 20,
  "digital_twin_update": 10
}

JOB_TIMEOUT

Type: int (seconds)
Default: 3600 (1 hour) Maximum execution time for a worker job before it’s terminated.

JOB_TIMEOUT=7200

Storage Plugin Configuration

STORAGE_PLUGIN_DIR

Type: string (directory path)
Default: /app/storage-plugins Directory where external storage plugins are cached and loaded from. The orchestrator compiles and loads .so files from this location.

STORAGE_PLUGIN_DIR=/app/storage-plugins

LLM Provider Configuration

These settings enable LLM-powered features like ontology generation and semantic extraction.

LLM_ENABLED

Type: bool
Default: false Enable LLM integration. When false, extraction uses statistical heuristics only.

LLM_ENABLED=true

LLM_PROVIDER

Type: string
Default: "" (empty = LLM disabled)
Values: openrouter | openai_compat | custom provider name LLM provider to use. Must be registered in the orchestrator.

LLM_PROVIDER=openrouter

Built-in providers:

Provider	Description
`openrouter`	OpenRouter multi-model API
`openai_compat`	OpenAI-compatible endpoints (e.g., vLLM, LocalAI)

LLM_API_KEY

Type: string (API key)
Default: "" API key for the LLM provider. Required for openrouter and most openai_compat providers.

LLM_API_KEY=sk-or-v1-xxxxxxxxxxxx

Store API keys in Kubernetes Secrets, not in plaintext ConfigMaps.

LLM_BASE_URL

Type: string (URL)
Default: "" Base URL for openai_compat providers. Not required for openrouter.

LLM_BASE_URL=http://localhost:8000/v1

Examples:

vLLM: http://vllm-server:8000/v1
LocalAI: http://localai:8080/v1
OpenAI: https://api.openai.com/v1

LLM_MODEL

Type: string
Default: Provider-specific (see below) Model identifier to use for LLM requests. When empty, defaults to:

openrouter: openrouter/free
openai_compat: gpt-4o-mini

LLM_MODEL=anthropic/claude-3.5-sonnet

LLM_PROVIDER_DIR

Type: string (directory path)
Default: /app/llm-providers Directory where external LLM provider plugins are cached and loaded from.

LLM_PROVIDER_DIR=/app/llm-providers

Database Configuration

DATABASE_URL

Type: string (connection string)
Default: "" (uses SQLite in STORAGE_DIR) Database connection string. When empty, the orchestrator uses SQLite at {STORAGE_DIR}/mimir.db.

DATABASE_URL=sqlite:///app/data/mimir.db

Currently, only SQLite is supported. PostgreSQL support is planned.

Frontend Configuration

Frontend Environment Variables

The frontend service uses the following variables:

PORT

Type: string
Default: 3000 HTTP port for the frontend server.

PORT=3000

API_URL

Type: string (URL)
Default: http://orchestrator:8080 URL of the orchestrator API. The frontend makes REST API calls to this endpoint.

API_URL=http://orchestrator:8080

In Docker Compose, this uses the service name orchestrator. In Kubernetes, use the Service DNS name:

API_URL=http://mimir-aip-orchestrator:8080

Complete Example Configurations

Docker Compose (Development)

docker-compose.yaml

services:
  orchestrator:
    environment:
      - ENVIRONMENT=development
      - LOG_LEVEL=debug
      - PORT=8080
      - STORAGE_DIR=/app/data
      - MIN_WORKERS=1
      - MAX_WORKERS=5
      - QUEUE_THRESHOLD=3
      - LLM_ENABLED=true
      - LLM_PROVIDER=openrouter
      - LLM_API_KEY=${OPENROUTER_API_KEY}
      - LLM_MODEL=openrouter/free

Kubernetes ConfigMap (Production)

apiVersion: v1
kind: ConfigMap
metadata:
  name: mimir-aip-config
data:
  ENVIRONMENT: "production"
  LOG_LEVEL: "info"
  PORT: "8080"
  STORAGE_DIR: "/app/data"
  MIN_WORKERS: "2"
  MAX_WORKERS: "50"
  QUEUE_THRESHOLD: "10"
  WORKER_NAMESPACE: "mimir-aip"
  WORKER_SERVICE_ACCOUNT: "mimir-worker"
  WORKER_IMAGE: "ghcr.io/mimir-aip/worker:0.1.1"
  WORKER_CPU_LIMIT: "4000m"
  WORKER_MEMORY_LIMIT: "8Gi"
  ORCHESTRATOR_URL: "http://mimir-aip-orchestrator:8080"
  LLM_ENABLED: "true"
  LLM_PROVIDER: "openrouter"
  LLM_MODEL: "anthropic/claude-3.5-sonnet"
---
apiVersion: v1
kind: Secret
metadata:
  name: mimir-aip-secrets
type: Opaque
stringData:
  LLM_API_KEY: "sk-or-v1-xxxxxxxxxxxx"
  WORKER_AUTH_TOKEN: "your-secure-random-token"

Kubernetes Deployment (with Secrets)

apiVersion: apps/v1
kind: Deployment
metadata:
  name: mimir-aip-orchestrator
spec:
  template:
    spec:
      containers:
      - name: orchestrator
        envFrom:
        - configMapRef:
            name: mimir-aip-config
        env:
        - name: LLM_API_KEY
          valueFrom:
            secretKeyRef:
              name: mimir-aip-secrets
              key: LLM_API_KEY
        - name: WORKER_AUTH_TOKEN
          valueFrom:
            secretKeyRef:
              name: mimir-aip-secrets
              key: WORKER_AUTH_TOKEN

Environment Variable Precedence

When the same variable is set in multiple locations:

Container env (highest priority)
Container envFrom (ConfigMap/Secret)
Dockerfile ENV
Application defaults (lowest priority)

Validation and Defaults

The orchestrator validates configuration on startup:

Missing required directories are created automatically (e.g., STORAGE_DIR)
Invalid numeric values fall back to defaults
Invalid log levels fall back to info
Missing LLM configuration disables LLM features gracefully

Check the orchestrator logs for configuration warnings:

docker compose logs orchestrator | grep -i "config\|warn"

Next Steps

Deploy with Docker Compose for local development
Deploy with Kubernetes for production
Configure LLM provider API keys for enhanced ontology extraction
Set up multi-cluster workers for scalability

Getting Started

Core Concepts

Deployment

Platform Features

MCP Integration

Advanced Topics

​Core Settings

​ENVIRONMENT

​LOG_LEVEL

​PORT

​STORAGE_DIR

​Worker Pool Configuration

​MIN_WORKERS

​MAX_WORKERS

​QUEUE_THRESHOLD

​WORKER_NAMESPACE

​WORKER_SERVICE_ACCOUNT

​WORKER_IMAGE

​WORKER_CPU_LIMIT

​WORKER_MEMORY_LIMIT

​Multi-Cluster Configuration

​CLUSTER_CONFIG_FILE

​WORKER_AUTH_TOKEN

​ORCHESTRATOR_URL

​Advanced Worker Settings

​WORKER_CONCURRENCY_LIMITS

​JOB_TIMEOUT

​Storage Plugin Configuration

​STORAGE_PLUGIN_DIR

​LLM Provider Configuration

​LLM_ENABLED

​LLM_PROVIDER

​LLM_API_KEY

​LLM_BASE_URL

​LLM_MODEL

​LLM_PROVIDER_DIR

​Database Configuration

​DATABASE_URL

​Frontend Configuration

​Frontend Environment Variables

​PORT

​API_URL

​Complete Example Configurations

​Docker Compose (Development)

​Kubernetes ConfigMap (Production)

​Kubernetes Deployment (with Secrets)

​Environment Variable Precedence

​Validation and Defaults

​Next Steps

Build docs developers (and LLMs) love

Core Settings

ENVIRONMENT

LOG_LEVEL

PORT

STORAGE_DIR

Worker Pool Configuration

MIN_WORKERS

MAX_WORKERS

QUEUE_THRESHOLD

WORKER_NAMESPACE

WORKER_SERVICE_ACCOUNT

WORKER_IMAGE

WORKER_CPU_LIMIT

WORKER_MEMORY_LIMIT

Multi-Cluster Configuration

CLUSTER_CONFIG_FILE

WORKER_AUTH_TOKEN

ORCHESTRATOR_URL

Advanced Worker Settings

WORKER_CONCURRENCY_LIMITS

JOB_TIMEOUT

Storage Plugin Configuration

STORAGE_PLUGIN_DIR

LLM Provider Configuration

LLM_ENABLED

LLM_PROVIDER

LLM_API_KEY

LLM_BASE_URL

LLM_MODEL

LLM_PROVIDER_DIR

Database Configuration

DATABASE_URL

Frontend Configuration

Frontend Environment Variables

PORT

API_URL

Complete Example Configurations

Docker Compose (Development)

Kubernetes ConfigMap (Production)

Kubernetes Deployment (with Secrets)

Environment Variable Precedence

Validation and Defaults

Next Steps