config.yml.example

# EfficientAI Configuration File

# Application Settings
app:
  name: "Voice AI Evaluation Platform"
  version: "0.1.0"
  debug: true
  secret_key: "your-secret-key-here-change-in-production"

# Server Settings
server:
  host: "0.0.0.0"
  port: 8000

# Database Configuration
database:
  url: "postgresql://efficientai:password@localhost:5432/efficientai"
  # Alternative: specify individual components
  # user: "efficientai"
  # password: "password"
  # host: "localhost"
  # port: 5432
  # db: "efficientai"

# Redis Configuration
redis:
  url: "redis://localhost:6379/0"
  # Alternative: specify individual components
  # host: "localhost"
  # port: 6379
  # db: 0

# Celery Configuration
celery:
  broker_url: "redis://localhost:6379/0"
  result_backend: "redis://localhost:6379/0"

# File Storage
storage:
  upload_dir: "./uploads"
  max_file_size_mb: 500
  blob_provider: s3  # s3 | gcs — single active cloud blob backend
  allowed_audio_formats:
    - "wav"
    - "mp3"
    - "flac"
    - "m4a"

# S3 Configuration (for data sources integration)
s3:
  enabled: false
  bucket_name: "your-s3-bucket-name"
  region: "us-east-1"
  access_key_id: "your-access-key-id"
  secret_access_key: "your-secret-access-key"
  endpoint_url: null  # Optional: for S3-compatible services (e.g., MinIO, DigitalOcean Spaces)
  prefix: "audio/"  # Prefix for audio files in bucket

# GCS Configuration (alternative to S3 when storage.blob_provider is gcs)
gcs:
  enabled: false
  bucket_name: "your-gcs-bucket-name"
  project_id: "your-gcp-project-id"
  credentials_path: null  # Optional path to service-account JSON; falls back to GOOGLE_APPLICATION_CREDENTIALS / ADC
  prefix: "audio/"  # Prefix for audio files in bucket (same layout as S3)

# Speaker Diarization (pyannote.audio)
# For accurate speaker identification in transcripts, configure pyannote.audio:
# 1. Create a HuggingFace account at https://huggingface.co
# 2. Accept model terms at https://huggingface.co/pyannote/speaker-diarization-3.1
#    and https://huggingface.co/pyannote/segmentation-3.0
# 3. Generate a read-access token at https://huggingface.co/settings/tokens
diarization:
  huggingface_token: null  # e.g., "hf_xxxxx"
  num_speakers: 2  # Force exact speaker count (2 = agent + customer). Set to null for auto-detect.

# CORS Settings
cors:
  origins:
    - "http://localhost:3000"
    - "http://localhost:8000"

# API Settings
api:
  prefix: "/api/v1"
  key_header: "X-API-Key"
  rate_limit_per_minute: 60

# -----------------------------------------------------------------------------
# Authentication
# -----------------------------------------------------------------------------
# EfficientAI supports three deployment models for authentication:
#
#   1. OSS self-hosted (default):
#        providers: [api_key, local_password]
#      -> Users sign up with email + password. API keys are minted from the
#         UI or via scripts/create_api_key.py for programmatic access.
#
#   2. Enterprise self-hosted (requires an EFFICIENTAI_LICENSE with oidc_sso):
#        providers: [api_key, external_oidc]
#      -> Humans sign in via your existing identity provider over OIDC:
#         Okta, Azure AD / Entra ID, Google Workspace, AWS Cognito, Auth0,
#         Ping, JumpCloud, OneLogin, etc. Machines still use API keys.
#         Drop local_password from the list to force SSO.
#
#   3. Cloud SaaS (requires oidc_sso + mfa_enforce, allow_signup: false):
#        providers: [api_key, external_oidc]
#      -> Signups happen through the billing flow, not the API.
#
# Providers listed here but not licensed (external_oidc) will return a 403
# with a pointer to the license feature that unlocks them.
auth:
  providers:
    - api_key
    - local_password
    # - external_oidc     # requires EFFICIENTAI_LICENSE feature: oidc_sso

  local_password:
    # Lifetime of the Bearer tokens minted at POST /auth/login (in minutes).
    # Keep this short; clients re-authenticate silently.
    token_ttl_minutes: 720  # 12 hours
    # Turn this off in Cloud SaaS to block self-serve signup.
    allow_signup: true

  # External OIDC (enterprise): bring your own IdP. Works with any
  # OIDC-compliant provider. See README.md > Authentication & Deployment
  # Recipes for copy-paste recipes for Okta, Azure AD, Google Workspace
  # and AWS Cognito.
  #
  # oidc:
  #   issuer: "https://example.okta.com"                    # REQUIRED
  #   audience: "efficientai"                               # expected `aud` claim
  #   client_id: "0oa..."                                   # SPA client id
  #   jwks_uri: "https://example.okta.com/oauth2/v1/keys"   # optional, derived from issuer
  #   default_org_name: "Example Inc"                       # used when no org claim is present
  #   org_claim_path: ["https://efficientai.com/org"]       # dotted path into the JWT to find the org name

# Judge Alignment (AlignEval-style hybrid integration).
# Lets users build LLM-as-a-judge evaluators backed by labeled data,
# measure alignment metrics (precision/recall/F1/Cohen's kappa) and
# optimize the judge prompt via the existing GEPA service.
# Per-org thresholds (min labels to evaluate / optimize) and the judge
# model selection are configured from the UI, NOT from this file -- the
# judge model dropdown is sourced from your configured AI Provider
# integrations.
judge_alignment:
  enabled: true            # operator-level kill switch
  csv_max_rows: 5000       # safety limit for CSV uploads

# Enterprise License (JWT signed with RS256). Unlocks gated features like
# oidc_sso, mfa_enforce, audit_export, voice_playground, gepa_optimization,
# call_imports.
# license:
#   key: "eyJhbGciOi..."