RabbittWatch AI Observability SDK

Install:

npm install @rabbittwatch/ai-observability

Configure:

RABBITTIZE_INGEST_KEY=rw_live_xxxxx
RABBITTIZE_ENDPOINT=https://rabbitize-api.rabbitt.ai
RABBITTIZE_SERVICE_NAME=support-api
RABBITTIZE_ENVIRONMENT=prod

Wrap a provider call:

import { RabbittWatchAI } from "@rabbittwatch/ai-observability";

const rwAI = new RabbittWatchAI({
  apiKey: process.env.RABBITTIZE_INGEST_KEY!,
  endpoint: process.env.RABBITTIZE_ENDPOINT,
  serviceName: process.env.RABBITTIZE_SERVICE_NAME || "app",
  environment: process.env.RABBITTIZE_ENVIRONMENT || "prod",
});

const response = await rwAI.trace(
  {
    name: "chat.answer",
    provider: "openai",
    model: "gpt-4o",
    endpoint: "/chat",
    input: messages,
  },
  () => openai.chat.completions.create({ model: "gpt-4o", messages })
);

Manual spans:

const trace = rwAI.startTrace({ name: "POST /tickets/classify", endpoint: "/tickets/classify" });
const span = trace.startSpan({
  name: "bedrock.classify",
  kind: "llm",
  provider: "bedrock",
  model: "anthropic.claude-3-5-sonnet-20241022-v2:0",
});

try {
  const result = await bedrockClient.send(command);
  span.end({ status: "success", promptTokens: 1800, completionTokens: 120, output: result });
} catch (error) {
  span.end({ status: "error", error });
  throw error;
} finally {
  await trace.flush();
}

By default, prompt and output previews are not sent. Enable captureInput or captureOutput only when your privacy policy allows it.