EvalLedger

EvalLedger is a web app (with optional CLI) that turns model evaluation into an auditable, repeatable release gate. Teams define evaluation suites (accuracy, latency, cost, safety, bias, regression tests) and run them automatically on every model or prompt change. Results are stored as immutable “evaluation receipts” tied to code commit, dataset snapshot, feature flags, and deployment environment. It supports LLM and classic ML workflows: offline batch evals, canary/AB online metrics, and red-team test packs. The product focuses on the unglamorous but painful part of MLOps: proving what changed, why it’s safe, and who approved it. It integrates with CI/CD and popular model registries, then produces a single shareable report for engineering, security, and compliance. Expect to compete with platforms that do more, but often feel heavy and expensive for mid-market teams.