Opik is an open-source platform for tracing, evaluating, and monitoring LLM apps and AI agents, with self-hosting options and production dashboards.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Opik is an open-source LLM observability and evaluation platform for AI agent tracing, LLM evaluation, prompt management, and production monitoring. It includes tracing, evaluation datasets and experiments, prompt management, online evaluation, agent optimization, and guardrails. It is Apache-2.0 licensed and可 self-hosted.
How it works
Opik logs LLM calls and traces, supports annotations via Python SDK or UI, and offers dashboards for production monitoring. It integrates with multiple frameworks and supports an SDK for tracing, evaluation, and CI/CD testing. The Quick Start shows installing the Python SDK and a @track decorator to log traces.
Getting started
Install the Python SDK and configure it:
pip install opik
opik configure
Wrap functions with the @track decorator to log traces:
from opik import track
@track
def my_function(input: str) -> str:
return input
For more setup options, see the Quickstart guide.
Recent releases
Latest releases (with dates):
- 2.2.18 (2026-08-04): fix: scope studio prompt editors to their message role; fix: trace optimizer L
- 2.2.17 (2026-08-04): Update automatically OpenAPI spec and Fern code; sync provider model definitions
- 2.2.16 (2026-08-04): dedup the feedback-scores trace_final CTE with argMax; docs
- 2.2.15 (2026-08-03): fix: name the real cause when an optimization run shows nothing usable; fix
- 2.2.13 (2026-07-31): Update model prices file; fix: wait on experiment-creation response in Playground smoke
Traction
GitHub stars: 21106; forks: 1671; open issues: 164
Behind the repo
Opik is developed by Comet and is Apache-2.0 licensed; it includes server backend, web app, tracing, datasets, experiments, evaluations, prompt management, online evaluation, and agent optimization components.
Caveats
License: Apache-2.0. Creation date: 2023-05-10. Last push: 2026-08-04. Topics include evaluation, llm-observability, llm-evaluation, prompt-engineering. Open issues: 164.