@cynsta/witness-node

Node.js SDK for sending Cynsta Witness trace events to the ingest API over OTLP/HTTP protobuf.

Install

npm install @cynsta/witness-node

For LangChain JS support:

npm install @cynsta/witness-node @langchain/core

Direct usage

import { CynstaTracer } from "@cynsta/witness-node";

const tracer = new CynstaTracer({
  ingestEndpoint: process.env.CYNSTA_INGEST_URL,
  apiKey: process.env.CYNSTA_INGEST_API_KEY,
  mode: "client_trust",
  serviceName: "my-worker",
});

tracer.recordLlmInteraction({
  model: "gpt-5.2",
  prompt: "Summarize this filing.",
  completion: "The filing describes...",
  provider: "openai",
  requestedModel: "gpt-5.2-chat-latest",
  resolvedModel: "gpt-5.2",
  tokensIn: 128,
  tokensOut: 64,
  temperature: 0.2,
});

await tracer.flush();
await tracer.shutdown();

LangChain JS

import { CynstaTracer } from "@cynsta/witness-node";
import { CynstaLangChainHandler } from "@cynsta/witness-node/langchain";

const tracer = new CynstaTracer({
  serviceName: "langchain-worker",
});

const handler = new CynstaLangChainHandler({ tracer });

// Pass `handler` into your LangChain runnable/model callbacks.

Environment variables

• CYNSTA_INGEST_URL
• CYNSTA_INGEST_API_KEY
• INGEST_API_KEY
• CYNSTA_MODE
• CYNSTA_USER_ID
• CYNSTA_PROMPT_RESPONSE_ENABLED
• CYNSTA_EXPORT_INTERVAL_MS
• CYNSTA_EXPORT_BATCH_SIZE
• CYNSTA_EXPORT_QUEUE_SIZE
• CYNSTA_EXPORT_TIMEOUT_MS

Notes

• Default transport is OTLP/HTTP protobuf to http://localhost:4318/v1/traces.
• client_trust mode emits cynsta.data, cynsta.sequence_index, cynsta.prev_hash, and cynsta.data_hash on every event.
• cynsta.config_json uses the same canonical shape as the Python tracer with kind: "llm_generation_config_v1".