jwcastillo / grafana-llmexamples-app

An example Grafana plugin showing how plugins might use LLMs from the Grafana frontend.

Grafana LLM examples

This is a Grafana plugin designed to showcase using the LLM functionality available in @grafana/experimental. Under the hood this uses the grafana-llm-app to proxy requests to the LLM provider.

Getting started

To get started you'll need a few things:

• an OpenAI key - see the 'Getting access to LLMs' section of the Getting Started with LLMs at Grafana doc
  • this should be made available as the OPENAI_API_KEY environment variable
• Docker

Then run the following to build the plugin:

npm install
npm run dev

Finally, use docker compose to run a Grafana instance with access to the plugin:

docker compose up

You should then be able to access Grafana on http://localhost:3000.

Head to the LLM Examples plugin page to see some use of the LLMs in action!

Explanation

The Grafana container in docker-compose is provisioned with the grafana-llm-app plugin installed (using GF_INSTALL_PLUGINS) and configured with your OpenAI key (using provisioning/plugins/apps.yaml).

This plugin makes use of the @grafana/experimental package to make requests to OpenAI via the grafana-llm-app plugin, which provides an authenticating proxy and handles streaming responses using Grafana Live.

Take a look at src/pages/ExamplePage.tsx to see how to make requests and use responses. You can also take a look at extensions/panelExplainer.tsx to see how to add a plugin extension utilizing the same APIs.

You can also toggle the value of disabled for the grafana-llm-app plugin in provisioning/plugins/apps.yaml to see what happens when the LLM plugin is unavailable.

Adapting your own plugin

To add LLM functionality to your own plugin you'll need to do the following:

• add a dependency on @grafana/experimental>=1.7.0 and make use of the llms module
• ensure the grafana-llm-app plugin is installed and configured in the Grafana instance
  • install by setting the GF_INSTALL_PLUGINS environment variable at startup time - see docker-compose.yaml for an example
  • configure either:
    • using provisioning (see provisioning/plugins/apps.yaml), or
    • within the app settings screen (e.g. http://localhost:3000/plugins/grafana-llm-app)