@profitstone/pstone

Profit Stone in-process Telemetry SDK for Node.js and TypeScript.

OpenTelemetry-based collection of LLM calls, sent to the Profit Stone SaaS Ingest API (same contract as the Python pstone package).

npm install @profitstone/pstone
# optional peers for auto-instrumentation:
npm install openai @anthropic-ai/sdk

import { pstone } from "@profitstone/pstone";
import OpenAI from "openai";

pstone.init(process.env.PSTONE_API_KEY!);

const client = new OpenAI();
const resp = await client.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [{ role: "user", content: "Hello" }],
});

await pstone.flushAsync();
await pstone.shutdownAsync();

0.2.0 — init / OTLP protobuf export / OpenAI & Anthropic instrumentation (parity with Python MVP).

Requirements

• Node.js 18+

Environment

| Variable | Required | Description | |----------|----------|-------------| | PSTONE_API_KEY | recommended | Workspace API Key if omitted from init | | PSTONE_INGEST_URL | optional | Ingest base URL (default https://api.profitstone.ai) | | PSTONE_SERVICE_NAME | optional | OTel service.name (default pstone-app) |

Traces are sent to {PSTONE_INGEST_URL}/v1/ingest/traces with Authorization: Bearer <key> (OTLP/HTTP protobuf).

API

| Method | Purpose | |--------|---------| | pstone.init(apiKey?, options?) | Start tracer + optional auto-instrument | | pstone.flush() / flushAsync() | Force export of pending spans | | pstone.shutdown() / shutdownAsync() | Uninstrument + shut down provider | | pstone.wrapOpenai(client) | Wrap one OpenAI client instance | | pstone.wrapAnthropic(client) | Wrap one Anthropic client instance |

init options

| Option | Default | Description | |--------|---------|-------------| | ingestUrl | env / SaaS default | Ingest base URL | | serviceName | pstone-app | Resource service.name | | captureContent | true | Include prompts/completions | | redact | true | Basic email/phone redaction | | instrument | true | Auto-patch OpenAI / Anthropic if installed | | instrumentOpenai | true | Patch OpenAI | | instrumentAnthropic | true | Patch Anthropic | | useBatch | true | BatchSpanProcessor (false for tests) | | exporter | OTLP protobuf | Inject custom exporter (tests) |

Instrumentation

With instrument: true (default), when packages are installed:

• OpenAI — chat.completions.create (+ parse / responses.create when present), including stream: true
• Anthropic — messages.create, including stream: true

Or wrap a single client: pstone.wrapOpenai(client).

Local smoke (optional)

export PSTONE_API_KEY=psk_...
export PSTONE_INGEST_URL=http://localhost:8000
node -e "
import { pstone } from '@profitstone/pstone';
import { startLlmSpan, recordLlmResponse } from '@profitstone/pstone';
pstone.init(undefined, { useBatch: false, instrument: false });
const s = startLlmSpan({ system: 'openai', model: 'smoke', prompt: 'hi' });
recordLlmResponse(s.span, { completion: 'ok', inputTokens: 1, outputTokens: 1 });
s.end();
await pstone.flushAsync();
await pstone.shutdownAsync();
console.log('flushed');
"

License

MIT