Skip to main content

​
Valiqor finds why your AI app fails

Most evaluation tools tell you something is wrong. Valiqor tells you what went wrong, why it happened, and how to fix it — with severity scoring, root-cause analysis, and actionable remediation.
You can get started in 5 minutes — no tracing or instrumentation needed. Just pass your existing AI inputs and outputs to Failure Analysis and see results immediately.

​
Five modules, one SDK

Valiqor’s Python SDK exposes five modules through a single client: 
from valiqor import ValiqorClient

client = ValiqorClient(api_key="vq_...")

client.failure_analysis   # Root-cause failure detection
client.eval               # Metric-based evaluation
client.security           # Red-team audits & vulnerability scanning
client.trace              # Auto-instrumentation for OpenAI, Anthropic, LangChain
client.scanner            # Repository & prompt scanning
ModuleWhat it doesWhen to use it
Failure AnalysisClassifies failures into buckets & subcategories, scores severity (0–5), explains root causeFirst thing to run — works on any input/output data
EvaluationRuns metric-based checks (hallucination, relevance, coherence, etc.)When you need granular metric scores
SecurityRuns red-team attacks, checks for prompt injection, data leakage, jailbreaksBefore deploying to production
TracingAuto-captures LLM calls, tool use, retrieval steps as structured tracesWhen you want continuous monitoring in production
ScannerScans repositories and prompts for best-practice violationsDuring code review or CI

​
Two ways to get started

Quick start — Run on existing data

Pass your AI inputs/outputs directly. No tracing, no instrumentation. See failure results in under 5 minutes.

Full observability — Add tracing

Instrument your LLM calls with auto-tracing. Run Failure Analysis on traces for continuous production monitoring.

​
Who is Valiqor for?

Valiqor is built for AI engineers and teams who build applications on top of:
  • LLM providers — OpenAI, Anthropic, Google, Mistral, Cohere, and any OpenAI-compatible API
  • Orchestration frameworks — LangChain, CrewAI, LlamaIndex, Haystack
  • Custom pipelines — RAG systems, agent loops, multi-step chains
Whether you’re prototyping or running in production, Valiqor helps you catch and fix failures before your users do.

​
What you get

1

Failure taxonomy

Every failure is classified into buckets (e.g., Hallucination, Context Ignorance) and subcategories (e.g., Entity Fabrication, Contradicts Source).
2

Root-cause analysis

Each failure includes a judge_rationale explaining exactly what went wrong, plus a scoring_breakdown with evidence.
3

Severity & confidence scoring

Severity from 0 (no issue) to 5 (critical). Confidence from 0.0 to 1.0. Filter and prioritize failures automatically.
4

Actionable remediation

Fix your prompts, retrieval pipeline, or guardrails based on concrete failure evidence — not guesswork.

Ready? See your first failure →

Run Failure Analysis on your existing data in under 5 minutes.