LangChain events, sealed as receipts.

Detailed trace

Use the callback handler for lifecycle receipts.

Tool start, tool end, agent action, LLM end, and errors can each become chain entries.

import "dotenv/config";
import { createRankigiCallback } from "@rankigi/langchain";
import { AgentExecutor } from "langchain/agents";

// Reads RANKIGI_API_KEY / RANKIGI_AGENT_ID / RANKIGI_PASSPORT_ID /
// RANKIGI_SIGNING_KEY from the environment and manages its own SDK.
const executor = new AgentExecutor({
  agent,
  tools,
  callbacks: [createRankigiCallback({ agentTag: "research-agent" })],
});

await executor.invoke({ input: "summarize this matter" });

Compact trace

Use wrap() for one receipt per run.

This is the safer default when reviewers need proof of the run but not every internal LangChain callback.

import "dotenv/config";
import { Rankigi } from "@rankigi/sdk";

const rankigi = Rankigi.fromEnv();

const input = { input: "summarize this matter" };
const result = await rankigi.wrap(
  "langchain-run",
  input,
  () => executor.invoke(input),
);

await rankigi.close();
console.log(result);