truthlayer-sdk 0.1.0

TruthLayer — AI Hallucination Verification SDK. One-line verification of AI outputs against source documents.

TruthLayer Python SDK

One-line AI hallucination verification. Zero dependencies.

Install

pip install truthlayer-sdk

With LangChain integration:

pip install truthlayer-sdk[langchain]

Quick Start

from truthlayer import TruthLayer

tl = TruthLayer(
    api_key="tl_your_key_here",
    api_url="https://qoa10ns4c5.execute-api.us-east-1.amazonaws.com/prod"
)

# Verify an AI response against source documents
result = tl.verify(
    ai_response="Refunds are processed within 5-7 business days.",
    source_documents=["Our refund policy allows returns within 30 days. Refunds are processed within 5-7 business days."]
)

print(f"Trust Score: {result.trust_score}%")
print(f"Claims: {result.total_claims}")
print(f"Hallucinations: {result.has_hallucinations}")
print(f"Latency: {result.latency_ms}ms")

API Reference

TruthLayer(api_key, api_url, timeout=30)

Initialize the client.

Parameter	Type	Required	Description
api_key	str	Yes	Your API key (starts with tl_)
api_url	str	Yes	Base URL of your TruthLayer deployment
timeout	int	No	HTTP timeout in seconds (default: 30)

tl.verify(ai_response, source_documents=None, document_ids=None)

Verify AI output against source documents.

Returns VerificationResult with:

Property	Type	Description
claims	List[Claim]	Individual claim verdicts
trust_score	float	Overall trust score (0-100)
has_hallucinations	bool	True if any unsupported claims
verified_count	int	Number of verified claims
unsupported_count	int	Number of unsupported claims
latency_ms	float	API response time

tl.upload_document(content, title="", metadata=None)

Upload a document for later verification by ID.

tl.get_document(document_id) / tl.list_documents() / tl.delete_document(document_id)

Document management endpoints.

tl.health()

Check API health status.

LangChain Integration

from truthlayer.langchain import TruthLayerOutputParser

parser = TruthLayerOutputParser(
    api_key="tl_xxx",
    api_url="https://your-api/prod",
    source_documents=["Your source text here."],
    min_trust_score=70.0,
)

# Use in a chain — blocks hallucinations automatically
chain = llm | parser
result = chain.invoke("What is our refund policy?")
print(result.trust_score)  # VerifiedOutput with trust score

Error Handling

from truthlayer import TruthLayer, TruthLayerError

tl = TruthLayer(api_key="tl_xxx", api_url="https://your-api/prod")

try:
    result = tl.verify("AI response", ["Source text"])
except TruthLayerError as e:
    print(f"API error: {e}")
except ValueError as e:
    print(f"Invalid input: {e}")

Get an API Key

Visit your TruthLayer dashboard or use the API directly:

import requests

response = requests.post(
    "https://qoa10ns4c5.execute-api.us-east-1.amazonaws.com/prod/keys",
    json={"owner": "Your Name", "email": "[e-mail]", "use_case": "Chatbot"}
)
key = response.json()["api_key"]  # Save this — shown once only

License

MIT