Ragas Examples

Example projects for evaluating different kinds of LLM applications with Ragas: RAG systems, agents, prompts, workflows, text-to-SQL and LLM benchmarking. They are the source of the ragas quickstart templates. They may change between releases.

Installation

The examples are the ragas-ng-examples package (import name ragas_examples), a member of this repository's uv workspace, published to PyPI alongside ragas-ng:

pip install ragas-ng-examples

To work on them, install from a clone:

git clone https://github.com/arpieb/ragas-ng.git
cd ragas-ng
uv sync --package ragas-ng-examples
# or, with pip, in editable mode:
pip install -e . -e ./examples

Some examples need extra dependencies, available as extras of ragas-ng-examples:

uv sync --package ragas-ng-examples --extra text2sql
# or: pip install -e "./examples[text2sql]"

Starting a project from a template

To copy an example into a new project instead of running it in place:

ragas quickstart                          # list the templates
ragas quickstart rag_eval -o ./my-project # copy one

Available examples

Standalone scripts in this folder:

• iterate_prompt/: the prompt iteration guide example.
• gdrive_backend_example.py, gdrive_append_example.py: datasets stored in Google Sheets. See Storage backends.
• oci_genai_example.py: Oracle Cloud Infrastructure Generative AI models through LiteLLM (llm_factory("oci/...")) scored with metrics_experiment(). See the OCI guide.

Running an example

Most examples call OpenAI, so set your key first. The llamaIndex_agent_evals example uses Gemini (GOOGLE_API_KEY). The quickstart guide shows how to point the rag_eval template at other providers.

export OPENAI_API_KEY=your_key_here

Each template names its models once, near the top of the module that uses them. Override them without editing code:

RAGAS_EXAMPLE_APP_MODEL=gpt-5.6-terra RAGAS_EXAMPLE_JUDGE_MODEL=gpt-6.1-sol python -m ragas_examples.rag_eval.evals

The app's client must reach the model you name. Most templates use an OpenAI client, so a non-OpenAI model also needs that provider's OpenAI-compatible endpoint (see the quickstart guide).

Run an example as a module, or with its console script:

python -m ragas_examples.rag_eval.evals
# or
ragas-rag-evals

Versioning

The package version comes from the repository's git tags through setuptools_scm. These are the same tags as the ragas package, so the two versions stay in step. The publish-examples.yml workflow builds and publishes the package when a GitHub release is published. This fork doesn't publish to PyPI.