Evidently AI

Evidently AI maintains this Python library for checking data, classic ML models, and LLM applications from notebook to production. You wrap a pandas DataFrame in a Dataset, attach descriptors that score each row (text length, sentiment, keyword hits, LLM-as-judge verdicts), and run a Report assembled from presets or individual metrics; adding threshold conditions turns it into a pass or fail Test Suite for CI. More than 100 checks cover drift (20-plus detection methods), missing values, classification and regression errors, NDCG-style ranking for retrieval, and recommender novelty. Output renders inline in Jupyter or exports as HTML, JSON, or a Python dict. Running evidently ui starts a self-hosted dashboard on port 8000 for tracking results over time, and the companion Tracely package captures OpenTelemetry traces. Install with pip or conda on Python 3.10 or later; no GPU is needed. The library is Apache-2.0. The docs say the hosted Evidently Cloud SaaS has ended, while a paid self-hosted enterprise edition adds no-code tools, alerts, and access control. ML engineers and LLM app teams are the audience.