NVIDIA NeMo Agent Toolkit

​

Prerequisites

• Python 3.11+
• uv installed
• An Arize AX account (sign up)
• The NeMo Agent Toolkit repository cloned locally

​

Launch Arize AX

1. Sign in to your Arize AX account.
2. From Space Settings, copy your Space ID and API Key. You will set them as ARIZE_SPACE_ID and ARIZE_API_KEY below.

​

Install

uv pip install "nvidia-nat[opentelemetry]"

​

Configure credentials

export ARIZE_SPACE_ID="<your-space-id>"
export ARIZE_API_KEY="<your-api-key>"
export ARIZE_PROJECT_NAME="simple_calculator"

​

Setup tracing

general:
  telemetry:
    tracing:
      arize_ax:
        _type: arize_ax
        space_id: ${ARIZE_SPACE_ID}
        api_key: ${ARIZE_API_KEY}
        project: ${ARIZE_PROJECT_NAME:-simple_calculator}
        # Optional: enable for EU data residency
        # use_eu_region: true

​

Run NeMo Agent Toolkit

uv pip install -e examples/observability/simple_calculator_observability/

nat run \
  --config_file examples/observability/simple_calculator_observability/configs/config-arize-ax.yml \
  --input "What is 2 * 4?"

​

Expected output

Started exporter 'arize_ax'
The result of 2 * 4 is 8.

​

Verify in Arize AX

1. Open your Arize AX space and select project simple_calculator (or whatever value you set for ARIZE_PROJECT_NAME).
2. You should see a new trace within ~30 seconds containing the workflow’s root span and nested LLM / tool spans with prompts, responses, and token usage attached.
3. If no traces appear, see Troubleshooting.

​

Check from the skill, CLI, or SDK

npx skills add Arize-ai/arize-skills

ax spans export "$ARIZE_PROJECT_NAME" --space "$ARIZE_SPACE_ID" \
  --limit 5 --stdout | jq 'length'

import os
from arize import ArizeClient

client = ArizeClient(api_key=os.environ["ARIZE_API_KEY"])
resp = client.spans.list(
    project=os.environ["ARIZE_PROJECT_NAME"],
    space=os.environ["ARIZE_SPACE_ID"],
    limit=5,
)
count = len(resp.spans)
print(
    f"{count} span(s) found" if count else "No spans yet — recheck setup"
)

// Reads ARIZE_API_KEY from the environment.
import { listSpans } from "@arizeai/ax-client";

const { data: spans } = await listSpans({
  project: process.env.ARIZE_PROJECT_NAME!,
  space: process.env.ARIZE_SPACE_ID!,
  limit: 5,
});
const count = spans.length;
console.log(
  count ? `${count} span(s) found` : "No spans yet — recheck setup",
);

client, err := arize.NewClient(
    arize.Config{APIKey: os.Getenv("ARIZE_API_KEY")},
)
if err != nil {
    log.Fatal(err)
}
resp, err := client.Spans.List(ctx, spans.ListRequest{
    Project: os.Getenv("ARIZE_PROJECT_NAME"),
    Space:   os.Getenv("ARIZE_SPACE_ID"),
    Limit:   5,
})
if err != nil {
    log.Fatal(err)
}
fmt.Printf("%d span(s) found\n", len(resp.Spans))

​

Troubleshooting

• No Started exporter 'arize_ax' log line. The OpenTelemetry extra isn’t installed in the active environment. Re-run uv pip install "nvidia-nat[opentelemetry]" and confirm the YAML _type is exactly arize_ax.
• No traces in Arize AX. Confirm ARIZE_SPACE_ID and ARIZE_API_KEY are exported in the same shell that runs nat run. The YAML uses ${ARIZE_SPACE_ID} / ${ARIZE_API_KEY} substitution, so unset variables resolve to empty strings and the exporter silently fails to authenticate.
• EU data residency. Set use_eu_region: true under general.telemetry.tracing.arize_ax to route traces to the EU ingest endpoint.
• Wrong project in Arize AX. The YAML project: value wins over ARIZE_PROJECT_NAME unless you use the ${ARIZE_PROJECT_NAME:-default} substitution shown above.

​

Resources