Configuration

Rdakt AI can be configured via a YAML file or programmatically in Python.

YAML Configuration

Create a rdakt.yaml file:

mode: active
on_error: warn_and_forward

pipeline:
  - regex

session:
  store: memory

entities:
  EMAIL:
    strategy: token
  PERSON:
    strategy: synthetic
  SSN:
    strategy: token

Load it in your code:

from rdakt_ai import RdaktMiddleware, load_config

config = load_config("rdakt.yaml")
middleware = RdaktMiddleware(config=config)

Configuration Options

`mode`

Value	Behavior
`active` (default)	Detect and anonymize PII
`audit`	Detect and log PII, but forward original text

`on_error`

Value	Behavior
`warn_and_forward` (default)	Log error, forward unanonymized
`block`	Raise exception, don't send request

`pipeline`

List of detectors to run, in order:

pipeline:
  - regex                    # Built-in regex patterns
  - ner:                     # spaCy NER (requires rdakt-ai[ner])
      model: en_core_web_sm

`entities`

Per-entity-type configuration:

entities:
  EMAIL:
    strategy: token          # Override strategy for this type
  EMPLOYEE_ID:
    pattern: "EMP-\\d{6}"   # Custom regex pattern
    strategy: token

`session`

Session store backend:

# In-memory (default)
session:
  store: memory

# SQLite
session:
  store: sqlite
  path: /var/lib/rdakt/sessions.db

# Redis
session:
  store: redis
  url: redis://localhost:6379
  ttl: 3600
  prefix: "rdakt:"

Programmatic Configuration

from rdakt_ai import RdaktMiddleware, RdaktConfig, EntityConfig, SessionConfig

config = RdaktConfig(
    mode="active",
    on_error="block",
    pipeline=["regex", {"ner": {"model": "en_core_web_sm"}}],
    entities={
        "EMAIL": EntityConfig(strategy="token"),
        "PERSON": EntityConfig(strategy="synthetic"),
        "EMPLOYEE_ID": EntityConfig(pattern=r"EMP-\d{6}", strategy="token"),
    },
    session=SessionConfig(store="sqlite", path="sessions.db"),
)

middleware = RdaktMiddleware(config=config)

Override at Middleware Level

The middleware constructor accepts mode and error_policy overrides that take precedence over the config:

config = load_config("rdakt.yaml")  # mode: active

# Override to audit mode for this specific middleware instance
middleware = RdaktMiddleware(config=config, mode="audit")

Optional. A list of per-field rules keyed by JSONPath, applied before the global pipeline so that structured fields (internal IDs, domain identifiers, addresses) get treated specifically rather than scanned as free text. Default is no ontology — global pipeline runs on every text field.

ontology:
  fields:
    - path: "$.messages[*].content"
      detect: ["EMAIL", "PHONE"]
    - path: "$.messages[*].metadata.user_id"
      replace_with_synthetic:
        type: USER_ID
        format: "user-{n:06d}"
    - path: "$.messages[*].context.case_number"
      detect_via_regex: '\bCASE-\d{4}-\d{6}\b'
      detect_via_regex_as: CASE_NUMBER
      mask_strategy: hash

See the Per-field PII ontology guide for the full rule reference, JSONPath subset, and validation rules.

Example Configurations

Minimal (defaults)

mode: active
on_error: warn_and_forward
pipeline:
  - regex
session:
  store: memory

Production

mode: active
on_error: block
pipeline:
  - regex
  - ner:
      model: en_core_web_sm
session:
  store: sqlite
  path: /var/lib/rdakt/sessions.db
entities:
  EMAIL:
    strategy: token
  PERSON:
    strategy: synthetic
  PHONE:
    strategy: token
  CREDIT_CARD:
    strategy: token
  EMPLOYEE_ID:
    pattern: "EMP-\\d{6}"
    strategy: token

Distributed (Redis)

mode: active
on_error: warn_and_forward
pipeline:
  - regex
session:
  store: redis
  url: redis://localhost:6379
  ttl: 3600
  prefix: "rdakt:"