The Validation Engine checks every agent action against your policies in real-time. Invalid decisions are blocked or escalated automatically. Sub-100ms latency.
Every agent action is validated synchronously before execution. P99 latency under 100ms even at scale.
Define rules in code or via our visual builder. Policies evaluate content, context, confidence, and custom logic.
Set minimum confidence thresholds. Actions below threshold automatically route to human review.
Ensure agent outputs conform to expected formats. Catch malformed data before it causes problems.
Validate meaning, not just structure. Detect inappropriate content, PII exposure, and policy violations.
Block harmful, offensive, or off-brand content automatically. Customizable to your specific requirements.
Add validation to your existing agent code with a single function call. Works with any LLM provider or agent framework.
from attest import Attest client = Attest(api_key="...") # Before executing, validate the action decision = client.validate( action="send_email", payload={ "to": recipient, "subject": subject, "body": agent_draft }, # Optional: include context for better decisions context={ "conversation_id": conv_id, "customer_tier": "enterprise" } ) if decision.approved: execute_action(decision.payload) elif decision.escalated: # Routed to human review await_review(decision.review_id) else: # Blocked by policy handle_rejection(decision.reason)
Integrate in minutes. First 1,000 validations free.