Opik
Opik is Comet's open-source LLM observability and evaluation platform under Apache-2.0, covering the whole repository with no enterprise folders. It traces agents with full trace trees, evaluates with datasets, experiments and LLM-as-a-judge metrics for hallucination and retrieval quality, manages prompts, monitors production with online evaluation rules, runs tests in CI through PyTest, and exposes an MCP server for coding agents. It self-hosts with Docker or Kubernetes.
What does Opik do?
Opik is "the open-source LLM observability and evaluation platform for AI agent tracing, LLM evaluation, prompt management, and production monitoring." Agent tracing records "full trace trees for multi-step agents and tool calls", and every call to a decorated function is logged with its nested calls. Evaluations use datasets, experiments and LLM-as-a-judge metrics for hallucination detection, moderation and retrieval assessment, plus heuristic metrics, and "a PyTest integration to test LLM pipelines on every commit."
Production dashboards, online evaluation rules, an agent optimiser and guardrails cover live systems, and a prompt playground and prompt management cover development. It has Python and TypeScript SDKs, a REST API and first-party OpenTelemetry support. Its integrations table has 64 rows, including Ollama, Anthropic, Bedrock, Gemini, Groq, Mistral, DeepSeek, LangChain, LlamaIndex, CrewAI and Spring AI. An MCP server works with Claude Code, Cursor, VS Code Copilot, Codex and opencode. The README says Opik is designed for more than 40 million traces a day, its own figure.
Key facts
- Licence: Apache-2.0 for the whole repository, including the server, web app, tracing, evaluations, prompt management, online evaluation and agent optimisation, by the README; no ee directory.
- Agent tracing with full trace trees for multi-step agents and tool calls.
- Evaluations: datasets, experiments, LLM-as-a-judge metrics for hallucination, moderation and retrieval, and heuristic metrics.
- CI testing through a PyTest integration; production dashboards, online evaluation rules, an agent optimiser and guardrails.
- SDKs for Python and TypeScript, a REST API, OpenTelemetry support, and 64 integrations including Ollama, Anthropic, Bedrock, LangChain and CrewAI.
- An MCP server for Claude Code, Cursor, VS Code Copilot, Codex and opencode, set up with uvx opik mcp configure.
- Install: ./opik.sh on Linux or macOS or opik.ps1 on Windows, which uses Docker Compose for local development; Kubernetes with Helm for production. Comet Cloud is the hosted option with a free account.
- Latest tagged release when read: 2.2.68 on 2026-09-17; 22,104 stars on 2026-09-17.
What does it replace, and where does it fall short?
Opik replaces LangSmith for a team that wants tracing, evaluations, prompt management and production monitoring under one permissive licence with no enterprise carve-outs, self-hosted, plus tests that run in CI. It is the pick for a single Apache-2.0 licence on the open-source alternatives to LangSmith page.
Where it falls short: the Docker Compose route is labelled for local development and testing, so production means Kubernetes. The 40 million traces a day figure is the project's own. Its README compares itself with other products, and those comparisons are not repeated here. For the most widely used option, Langfuse.
How does Opik run?
Clone the repository and run ./opik.sh on Linux or macOS or opik.ps1 on Windows, which starts Docker Compose with the interface at localhost:5173, for local development. For production use Kubernetes with Helm. The client installs with pip install opik and opik configure. Comet Cloud is the hosted option, which the README calls the easiest, with a free account.
Who is Opik for?
A team that wants the whole observability and evaluation stack under Apache-2.0, including production monitoring and CI tests, self-hosted. A team already on OpenTelemetry across many languages might compare Phoenix.
What limits does the README state?
From the README: the Docker Compose route is for local development and testing; ./opik.sh --clean deletes all Opik data; missing features should be raised as requests.
Questions people ask
Is Opik open source?
Yes. The README says the server, web application, tracing, datasets, experiments, evaluations, prompt management, online evaluation and agent optimisation components are all under Apache-2.0, and the repository has no enterprise directory. Comet also offers a hosted version.
Can Opik test LLM applications in CI?
Yes. It has a PyTest integration to test LLM pipelines on every commit, alongside datasets, experiments and LLM-as-a-judge metrics for hallucination, moderation and retrieval quality, so a regression in answer quality can fail a build like any other test.
Does Opik work with local models?
Yes. Ollama is in its integrations table of 64 entries, alongside hosted providers such as Anthropic, Bedrock, Gemini and Groq and frameworks such as LangChain, LlamaIndex and CrewAI. It also accepts OpenTelemetry data.
How does Opik compare with LangSmith?
Both trace, evaluate and manage prompts for LLM applications. Opik self-hosts under a single Apache-2.0 licence, adds CI tests and production monitoring, and works across frameworks. The open-source alternatives to LangSmith page compares it with Langfuse and Phoenix.
Can coding agents use Opik?
Yes, through its MCP server, which the README says works with Claude Code, Cursor, VS Code Copilot, Codex and opencode, set up with uvx opik mcp configure. Agents can then read traces and evaluation results.
Sources
- Opik README and Apache-2.0 LICENSE: github.com/comet-ml/opik, read 2026-09-17.
Compared with the others
On the open-source alternatives to LangSmith page, Opik is the pick for one apache-2.0 licence, plus ci tests. Agent tracing, LLM-as-a-judge evaluations, prompt management, production monitoring and a PyTest integration, all Apache-2.0 with no enterprise folders.
Also on that page: Langfuse for the most widely used, mit core, Arize Phoenix for opentelemetry across four languages.
More llm observability and evaluation
An open-core LLM engineering platform, MIT outside its ee folders and part of ClickHouse since January 2026, for tracing, prompt management, evaluations, datasets and a playground, with Python and TypeScript SDKs.
Arize's AI observability platform built on OpenTelemetry for tracing, evaluations, datasets, experiments and prompt management, under the Elastic License 2.0, which forbids offering it as a hosted service.
Added September 17, 2026. Every claim above comes from the project's README, LICENSE or model card, read on September 17, 2026, or from the GitHub API on the date shown in the panel. Found an error? Write to reveneau@licheo.com and it is fixed in the next weekly pass. Repository: github.com/comet-ml/opik.
Running one of these inside your own environment, with your own data and your own security rules, is the kind of work Reveneau does. Read how a forward deployed engagement works.