> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://www.comet.com/docs/opik/llms.txt.

**Opik Dashboard**:

* Added the option to download traces and LLM calls as CSV files from the UI:

  ![](/docs/opik/_fern-files/_img/09d641387b2b068e45b39b4d66da84360d912ad8fe7e7106527ebec337185e6e.webp)
* Introduce a new quickstart guide to help you get started:

  ![](/docs/opik/_fern-files/_img/3d8036f8df332e990b58a89a5d0c3a065ec6d71be9a39413d763b6c4ee0798c8.webp)
* Updated datasets to support more flexible data schema, you can now insert items with any key value pairs and not just `input` and `expected_output`. See more in the SDK section below.
* Multiple small UX improvements (more informative empty state for projects, updated icons, feedback tab in the experiment page, etc).
* Fix issue with `\t` characters breaking the YAML code block in the traces page.

**SDK**:

* Datasets now support more flexible data schema, we now support inserting items with any key value pairs:

  ```python wordWrap
  import opik

  client = opik.Opik()
  dataset = client.get_or_create_dataset(name="Demo Dataset")
  dataset.insert([
      {
          "user_question": "Hello, what can you do ?",
          "expected_output": {
              "assistant_answer": "I am a chatbot assistant that can answer questions and help you with your queries!"
          }
      },
      {
          "user_question": "What is the capital of France?",
          "expected_output": {
              "assistant_answer": "Paris"
          }
      },
  ])
  ```

* Released WatsonX, Gemini and Groq integration based on the LiteLLM integration.

* The `context` field is now optional in the [Hallucination](/docs/opik/integrations/overview) metric.

* LLM as a Judge metrics now support customizing the LLM provider by specifying the `model` parameter. See more in the [Customizing LLM as a Judge metrics](/docs/opik/evaluation/metrics/overview#customizing-llm-as-a-judge-metrics) section.

* Fixed an issue when updating feedback scores using the `update_current_span` and `update_current_trace` methods. See this Github issue for more details.