Comet

Comet is a long-standing ML platform that has evolved into a unified observability layer for both classical ML and modern LLM applications. The platform spans the full lifecycle: Experiment Management tracks every training run with hyperparameters, metrics, code, data, and reproducible artifacts; the Model Registry catalogs trained models with lineage and approvals; Production Monitoring watches deployed models for drift, degradation, and data quality issues; and Opik (Comet's open-source LLM evaluation tool) provides LLM-specific tracing, evals, prompt management, and a playground. Opik is notable for being fully open source under Apache 2.0 with a hosted cloud option, giving teams a way to instrument LLM apps without sending traces to a closed platform. Engineering and research teams use Comet to keep dozens or hundreds of concurrent experiments organized, share results with stakeholders via reports, and avoid the classic 'which run produced this model?' problem. For LLM applications, Opik tracks every prompt and completion, scores outputs with code-based or LLM-as-judge evaluators, and helps catch regressions before they reach users. Comet has been around long enough to integrate with every major ML framework (PyTorch, TensorFlow, scikit-learn, XGBoost, Hugging Face, LangChain, LlamaIndex). Pricing offers a free tier for individuals, a paid Team tier, and Enterprise plans with on-prem deployment options.