anthropic

Anthropic AI Monitoring Example

A simple Fastify application demonstrating New Relic AI monitoring with the Anthropic SDK.

Setup

cp env.sample .env

Fill in your ANTHROPIC_API_KEY, ANTHROPIC_BASE_URL (if needed), NEW_RELIC_LICENSE_KEY, and NEW_RELIC_HOST (if needed) in .env, then:

npm ci
npm start

Endpoints

Chat Completion

curl -X POST http://localhost:3000/chat-completion \
  -H 'Content-Type: application/json' \
  -d '{"message": "Write a haiku about coding"}'

Streaming Chat Completion

curl -X POST http://localhost:3000/chat-completion-stream \
  -H 'Content-Type: application/json' \
  -d '{"message": "Write a short story about a developer who discovers their code has become sentient"}'

Error Endpoints

Endpoints for testing error instrumentation. Each triggers a different Anthropic SDK error type:

• UnprocessableEntityError (422) — invalid model name

  curl -X POST http://localhost:3000/error/unprocessable

• APIConnectionError — non-routable host (will timeout)

  curl -X POST http://localhost:3000/error/connection

Record Feedback

Use the id from a chat completion response:

curl -X POST http://localhost:3000/feedback \
  -H 'Content-Type: application/json' \
  -d '{"id": "<response_id>", "rating": 5, "message": "Great response"}'

Inspecting in New Relic

Navigate to AI Monitoring in your New Relic account to see AI response data, traces, and feedback events.