Comet

Comet builds infrastructure that development teams use to create, evaluate and improve AI applications and move prototypes into production systems. Its flagship open-source product, Opik, provides LLM observability and evaluation for generative AI applications and agentic systems: traces capture each step an agent takes, from context retrieval and tool selection through user feedback scores, and can be instrumented through more than 60 integrations or accessed by coding agents via an MCP server. Opik's Diagnostics feature surfaces errors from large volumes of traces, groups recurring issues, identifies root causes and recommends fixes even absent explicit error messages, while an embedded agent called Ollie reads traces and proposes code-level fixes to tool calls, retrieval steps and system prompts.

Beyond tracing and debugging, the platform offers Test Suites for pass/fail scoring of trace sets against defined goals, golden datasets and more than 40 LLM-as-a-judge evaluation metrics, human annotation by expert reviewers, and automated prompt engineering and optimization for multi-step agentic systems. Production capabilities include dashboards, alerting, model cost tracking and governance support, plus Cost Intelligence for monitoring engineering teams' spend on coding agents such as Claude Code and Codex across MCP installs, skills, model selection, context retrieval and configuration.

Comet also retains its earlier MLOps product line, Experiment Management, which tracks and compares model training runs with custom visualizations, model versioning, dataset management (including Comet Artifacts) and production monitoring, and integrates with frameworks such as PyTorch, PyTorch Lightning, Keras, Scikit-learn, TensorFlow, Spark NLP, Amazon SageMaker, Google Vertex AI, GitLab and New Relic.

Founding story

Comet launched in 2017. CEO and co-founder Gideon Mendels is a computer scientist and ML researcher who previously co-founded GroupWize, where NLP models were trained and deployed to process billions of chats, and worked on hate speech and deception detection at Columbia University and Google. CTO and co-founder Nimrod Lahav is a computer scientist and software engineer who held leadership and engineering roles at Wix, VMware and Amdocs before co-founding GroupWize, whose team deployed more than 50 NLP models in 15 languages; he holds a bachelor's degree from the Academic College of Tel Aviv, Yaffo.

Business model

Comet sells a developer platform on a freemium basis: the Opik LLM observability and evaluation layer is available as open source and can be self-hosted, tried in Comet's cloud, or deployed under custom enterprise arrangements. Accounts are free to create with no credit card and a free tier intended for ongoing use, and the platform is free for individual users, early-stage startups and academic teams, with paid enterprise offerings sold through a sales team and demos. Deployment options include on-premises, cloud and hybrid.

Revenue comes from paid subscriptions to the commercial Comet platform and enterprise deployments layered on a free tier and an open-source core; the company reported 5X annual recurring revenue growth in the 12 months to November 2021.

Traction

Comet reports more than 150,000 users, over 10,000 teams, 22,000+ GitHub stars for Opik and 450 enterprise, startup and academic teams, alongside $70M in funding and 80 employees in 14 countries. In November 2021 the company reported over 150 customers, including Ancestry, Cepsa, Etsy, Uber, WorkFusion and Zappos, 5X ARR growth and a 3X increase in its global workforce over the prior 12 months; Zappos cited a Comet-supported ML model that reduced size-related order returns by about 10%.

Latest developments

Recent platform work centers on connecting observability to remediation: Opik's Diagnostics detects and groups recurring trace issues and recommends fixes via an MCP server, the Ollie agent proposes and tracks fixes inside Opik, Test Suites and LLM-as-a-judge evaluations validate changes, and Cost Intelligence gives engineering leaders visibility into Claude Code and Codex usage and token waste. Comet reports 22,000+ GitHub stars, 150,000+ users and 10,000+ teams on the platform.