Opik
Comet's open-source platform for tracing, evaluating and monitoring LLM apps and agents, with a self-hostable backend.
Comet's full platform, backend included, free to self-host with Docker or Kubernetes.
Opik is Comet's LLM observability and evaluation platform, and the repository contains the server backend and web app, not just an SDK. It suits ML and agent teams who want an Apache-2.0 LangSmith alternative they can run in their own environment.
What it is
Opik traces LLM calls and multi-step agents as full trace trees, then adds datasets, experiments and LLM-as-a-judge metrics for hallucination, moderation and RAG assessment. It also ships an Agent Optimizer SDK, guardrails, online evaluation rules, production dashboards and a PyTest integration for running evals on every commit.
Who it's for
It suits teams moving agents from prototype to production who want evals, monitoring and prompt optimization in one self-hostable product. It is heavier to run than a tracing library, so a solo developer who just wants quick logs may find a lighter tool easier.
Key features
Agent trace trees
Capture LLM calls, tool executions and retrieval steps as nested traces.
LLM-as-a-judge metrics
Built-in metrics for hallucination detection, moderation and RAG assessment.
PyTest eval integration
Run evaluations in CI so each commit checks your LLM pipeline.
Agent Optimizer SDK
Use the optimizer to improve prompts and agents.
Online evaluation rules
Score production traces automatically and watch feedback scores, trace counts and token usage on dashboards.
Coding-agent MCP setup
Run uvx opik mcp configure so Claude Code, Cursor, Codex and others can read traces and run evaluations.
Up and running in 3 steps
Our honest verdict
What's great
- The whole platform, backend and web app included, is in the Apache-2.0 repository, so self-hosting is not a cut-down edition.
- A @track decorator is enough to log nested calls, as shown in the Python quickstart.
- The README says it is designed for scale, citing 40M+ traces a day.
- Evaluation, guardrails and optimization sit in the same product as tracing, so you do not stitch several tools together.
Where it falls short
- The Docker Compose route is described as for local development on one machine; for scale the README sends you to Kubernetes with Helm.
- The stack starts databases and caches and several containers (opik.sh has infra, backend and full modes), so you need Docker and enough machine to run them.
- Some features, such as the Ollie assistant and coding harness, appear on the pricing page as cloud plan items with a trial then paid tokens, not as part of the free self-hosted package.
- The README includes its own comparison table against competitors, so treat its ratings of rival tools as the vendor's view.
Pricing & license
The code is free under Apache-2.0 and the pricing page lists an Open Source option at $0. Comet's hosted plans (comet.com/site/pricing) are Free Cloud at $0 with up to 10 team members, 25k spans a month and 60-day retention; Pro Cloud at $19 a month with up to 50 members, 100k spans and extra spans at $5 per 100k; and Enterprise on custom terms with SSO and flexible deployments.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Is Opik fully open source?
Yes, the repository is Apache-2.0 and the README says the server, web app and core observability and evaluation features can be self-hosted without a commercial licence.
Does Opik only work with specific frameworks?
No, the README says it is framework-agnostic and supports its SDK, OpenTelemetry and framework-specific integrations.
How do I run it for a team?
Use the Comet cloud, or self-host with Kubernetes and Helm. The Docker Compose script is described as for local development on one machine.
