agents · test cases · scans run
Normal scan · Change mode

Scan History

ScanWorkflowStatus CurrentCandidateChange ResultTime

Workflows

Score Over Time

Current score over successive scans

What Works

Failure Analysis

Failure Clusters by Frequency

Run a scan to see failure clusters

Select a cluster to view details

Rubrics

Loading workflow...

Rubric Library

Connect Your Agents
Waiting for traces

Point your existing OpenTelemetry exporter at Spectral. One environment variable, zero code changes. Spectral is never in the hot path — if we go down, your agents keep running.

Environment Variables
export OTEL_EXPORTER_OTLP_ENDPOINT=https://api.runspectral.com
export OTEL_EXPORTER_OTLP_HEADERS="X-Spectral-Key=sk-spectral-your-key-here"
Python (auto-instrument)
# pip install opentelemetry-instrumentation-langchain
from opentelemetry.instrumentation.langchain import LangchainInstrumentor
LangchainInstrumentor().instrument()  # That's it. All spans flow to Spectral.
Works with LangChain, LlamaIndex, CrewAI, OpenAI Agents SDK, Vercel AI SDK, and any OTel-compatible framework.
Developer Tools

Case Explorer

— PASS — FAIL
Case IDAgentConfigStatusScoreFailure Clusters

Experiments

Show scoring weights
Objective: Accuracy 1.0× · Hallucination 5.0× · Tail Risk 10.0× · Cost 0.2×

Deploy

Validation Suite

Run the 3-case proof pack: Weak Rescue, Decent Lift, False Defense

Run the validation suite to generate a trust report

Tests 3 scenarios: weak agent rescue, decent agent lift, and false improvement defense