NeMo Guardrails Integration

NeMo Guardrails uses Colang flows to constrain LLM behavior. Register CortexDB as custom actions so flows can recall context and capture audit-relevant turns.

Install

pip install cortexdbai nemoguardrails

Register custom actions

# config/actions.py
import os
from datetime import datetime, timezone
from uuid import uuid4
from nemoguardrails.actions import action
from cortexdb.v1 import V1Client

client = V1Client(api_url="https://api-v1.cortexdb.ai", actor="user:alice",
                  bearer=os.environ["CORTEX_TOKEN"])
SCOPE = "org:acme/user:alice"


@action()
async def cortex_recall(query: str) -> str:
    pack = client.recall(scope=SCOPE, view="holistic", query=query,
                         include=["beliefs", "facts", "episodes"],
                         budgets={"max_tokens": 3000})
    return pack["context_block"] or ""


@action()
async def cortex_capture(text: str, role: str = "user") -> bool:
    client.experience(scope=SCOPE, text=text, role=role,
                      observed_at=datetime.now(timezone.utc).isoformat(),
                      idempotency_key=f"nemo-{role}-{uuid4()}")
    return True

# config/rails.co
define flow recall_and_answer
  user said something
  $context = execute cortex_recall(query=$user_message)
  bot $context  # prepend to system
  execute cortex_capture(text=$user_message, role="user")

The forget surface (/v1/forget, /v1/erasures) is useful when guardrails detect a violation — call client.forget(...) from a block_message flow to drop the offending turn from derived layers.

See also

• Python SDK
• Forget