Install the SDK and capture a generation

This milestone happens on your machine, in the shell from Milestone 4. Install the SDK, copy the capture sample from your language page, and change only how the client is created so it reads those variables.

1. Run the install command for your language in the shell from Milestone 4.

   Expand table

   Language	Install command
   Python	pip install agento11y
   TypeScript and JavaScript	npm install @grafana/agento11y
   Go	go get github.com/grafana/agento11y/go
   Java	Add the com.grafana.agento11y:agento11y-core Maven dependency. The Java install snippet does not state a version, so check Maven Central for the current agento11y-core version before you add it.
   .NET	dotnet add package Grafana.Agento11y and dotnet add package Grafana.Agento11y.OpenAI

2. In the following table, click the link for the language you are using and locate the code sample.

   Expand table

   Language	Section
   Python	Capture a generation manually
   TypeScript and JavaScript	Capture a generation manually
   Go	Capture a generation
   Java	Capture a generation
   .NET	Capture a generation

3. Copy the code sample into a file you can run.

   The Python, TypeScript and JavaScript, and Go samples run as copied. The Java sample is a fragment, so wrap it in a class with a main method and add imports for the com.grafana.agento11y types it uses. The .NET sample uses an openAIClient and inputItems it doesn’t define, so create an OpenAI client with your provider key and a list of input items before the OpenAIRecorder.CreateResponseAsync call.

4. Change how the client is created so it reads the variables from Milestone 4.

   Find the lines that create the client. They include the endpoint. Replace them with the one line for your language in the following table, which is also under Use environment variables instead on the same page. Leave every other line as it is, including the shutdown call.

   For example, in Python, replace:

   Python Copy

   client = Client(
       ClientConfig(
           generation_export_endpoint="{AGENTO11Y_ENDPOINT}",
       )
   )

   with:

   Python Copy

   client = Client()

   The import line and the rest of the sample stay as they are. ClientConfig in the import is unused after this change, and leaving it there is fine.

   Expand table

   Language	Client
   Python	client = Client()
   TypeScript and JavaScript	const client = new Agento11yClient();
   Go	client := agento11y.NewClient(agento11y.Config{})
   Java	Agento11yClient client = new Agento11yClient();
   .NET	var client = new Agento11yClient();

5. Run the file from the shell where you set the variables.

   The Python, TypeScript and JavaScript, Go, and Java samples don’t call an LLM provider, so no provider key is needed. Each records one generation: conversation conv-1, model openai/gpt-4o, and the assistant message Hello.

   The .NET sample calls the OpenAI provider helper, so run it with your provider key.

The SDK for your language is installed, and you have captured a generation.