Verify by kluster.ai is an intelligent agent that flags hallucinations and factual errors in real‑time, so every response your model returns is backed by evidence.
Works with any LLM, needs zero threshold‑tuning, and integrates via a single API.
from os import environ
from openai import OpenAI
from getpass import getpass
# Get API key from user input
api_key = environ.get("API_KEY") or getpass("Enter your kluster.ai API key: ")
print(f"Sending a reliability check request to kluster.ai...\n")
# Initialize OpenAI client pointing to kluster.ai API
client = OpenAI(
api_key=api_key,
base_url="https://api.kluster.ai/v1"
)
Effortless, OpenAI-Compatible Setup
Easily integrate into your existing workflow with a quick base-URL swap—point your OpenAI client at kluster.ai and start auto-verifying responses in minutes.
Continuous, Real‑Time Validation
Verify operates in the critical path, catching hallucinations before they reach users and returning explanations your team can act on—so you ship features, not apologies.
{
"is_hallucination": true,
"usage": {
"completion_tokens": 154,
"prompt_tokens": 1100,
"total_tokens": 1254
},
"explanation": "The response provides a wrong location for the Eiffel Tower.\nThe Eiffel Tower is actually located in Paris, France, not in Rome.\nThe response contains misinformation as it incorrectly states the tower's location.",
"search_results": []
}
Transparent by design—insights and feedback in one place.
Inline Citations & Rationales
Every verdict ships with the sentences, URLs, or doc IDs that back it up—plus a short “reason” field your app can surface to users or analysts. Instant context means faster fixes and higher end‑user trust.
Audit‑Ready Logs
Per‑request verdicts stored in your cloud for SOC 2 / HIPAA reviews.
Yes. Verify is model‑agnostic; pass your prompt + response (or context docs) via a single REST/SDK call. No retraining or wrappers required.
Do I need to tune thresholds or label data first?
No. Verify ships with sensible defaults and begins catching hallucinations immediately — zero manual configuration.
How accurate is it?
Across 25 k + benchmark samples Verify delivered 11 % higher overall accuracy and better precision than CleanLab TLM while matching sub-10 s response times.
What about false positives?
The engine is optimized for high precision, dramatically reducing false alarms and unnecessary manual reviews.
Is our data secure and compliant?
Every deployment runs with disk encryption, and SOC 2 Type II controls; we never log your prompts or weights.
Which integrations are available out of the box?
REST & OpenAI-compatible APIs, plus ready-made connectors for n8n, Dify, MCP-server apps and more, so you can drop Verify straight into existing workflows.
