Changelog

What's new

All the latest features, improvements, and fixes shipped to EvalGuard.

v0.3.0

Latest

March 6, 2025

feature50+ adversarial attack templates covering OWASP LLM Top 10

featureMulti-turn automated red teaming with escalation strategies

featureCustom attack scenario YAML DSL

featureOWASP LLM Top 10 compliance reports with remediation guidance

improvementSecurity scan speed improved by 3x with parallel execution

improvementPII detection now covers 40+ entity types across 12 languages

fixFixed false positive in prompt injection detection for code blocks

fixFixed compliance report PDF generation timeout for large scans

February 1, 2025

featureFull agent trace visualization with interactive timeline

featureInfinite loop and cycle detection in agent execution

featureAI Gateway with semantic caching (67% avg hit rate)

featureSmart routing across OpenAI, Anthropic, Google, Mistral

improvementDashboard load time reduced by 60% with streaming SSR

improvementEvaluation results now include confidence intervals

fixFixed trace tree rendering for deeply nested agent calls (>20 levels)

fixFixed race condition in concurrent evaluation batch processing

January 10, 2025

feature14 built-in evaluation scorers (faithfulness, relevance, toxicity, etc.)

featureCustom LLM-as-judge evaluators with any grading rubric

featureDataset management with golden test sets and versioning

featureBasic security scanning with 12 attack templates

featureCI/CD integration via GitHub Actions and CLI

featureTeam dashboards with evaluation history and trends

featurePython and TypeScript SDK with full type safety

featureREST API with OpenAPI specification