Real-time verification of AI outputs against source documents. The invisible trust layer between your LLM and your users.
Powered by Enterprise-Grade Infrastructure
Pass any AI-generated response and the source documents through our API or SDK.
Semantic embeddings + entity contradiction detection verify each claim independently.
Each claim gets a confidence score and classification: Verified, Uncertain, or Unsupported.
This is a real API call to our live production endpoint — not a mockup.
Fully serverless. Built entirely on AWS.
Verify medical AI outputs against clinical guidelines and drug databases. Prevent dangerous misinformation.
Check contract analysis and legal research against source statutes. Ensure compliance accuracy.
Validate financial analysis against reports and filings. Catch fabricated statistics before they mislead.
From install to first verification in under 2 minutes.
pip install truthlayer-sdk
from truthlayer import TruthLayer
tl = TruthLayer(api_key="tl_your_key_here")
result = tl.verify(
ai_response="Python 3.11 is 25% faster.",
source_documents=["Python 3.11 has up to 25% speedup."]
)
for claim in result.claims:
print(f"{claim.status}: {claim.text} ({claim.confidence}%)")
# VERIFIED: Python 3.11 is 25% faster. (91.5%)Live Bedrock verification in ~900ms. Cached responses even faster at ~750ms.
Zero false alarms — two-signal verification with entity contradiction detection.
Monitor hallucination rates, trust scores, and verification trends live.
pip install truthlayer-sdk → result = tl.verify(response, sources)
DynamoDB embedding cache — 100% hit rate on repeated content, 1.4x speedup.
Catches numerical, negation, and superlative contradictions that embeddings miss.
Start free. Scale as you grow.
For developers exploring AI verification
For teams deploying AI in production
For organizations with compliance needs