Opik is an open-source platform by Comet for debugging, evaluating, and monitoring LLM applications, RAG systems, and agentic workflows. It has earned more than 22,700 stars on GitHub, is actively maintained with updates in October 2026, and is licensed under Apache-2.0.
Opik gives teams full visibility into what their AI applications actually do in development and production. It records comprehensive traces of every LLM call, retrieval step, and agent action, then layers automated evaluations on top so quality problems surface before users notice them. Production-ready dashboards track cost, latency, and quality over time, and a prompt playground plus prompt management tools help teams iterate on prompts with data instead of guesswork.
Opik is designed to cover the whole lifecycle of an LLM feature. During development, traces make it obvious why an agent took a wrong turn or a retrieval step returned junk. Before release, teams run structured experiments over shared datasets and compare candidate prompts and models side by side. In production, the same tracing continues, with online evaluation scoring live traffic so quality drift, rising hallucination rates, or cost spikes are caught early. Annotation queues let human reviewers label failures and feed them back into test datasets, closing the loop between real-world behavior and the next iteration.
Opik’s open-source edition is free to self-host. Comet also offers Opik Cloud, a managed version with a free tier and paid plans for larger teams and higher trace volumes, making the overall offering freemium.
Opik is built for AI engineering teams shipping LLM applications to production who need reliable evaluation, observability, and a repeatable way to catch regressions in prompts, models, and retrieval pipelines.