tp.wrapToolCall() method — TracePilot SDK reference

tp.wrapToolCall() wraps any asynchronous function — a database write, an email send, a web search, or any other tool invocation — and records it as a structured span linked to a parent span in your execution tree. Set isDestructive: true on operations that modify external state so the dashboard surfaces a warning badge, making it easy to audit side effects during debugging.

Signature

tp.wrapToolCall<T>(
  toolName: string,
  call: () => Promise<T>,
  parentSpanId: string,
  stepOrder: number,
  isDestructive?: boolean
): Promise<{ result: T; spanId: string }>

Parameters

string

required

A human-readable display name for the tool. This label appears in the dashboard’s execution tree and is used to identify the span when debugging.

() => Promise<T>

required

A zero-argument async function that executes the tool. Wrap your call in an arrow function so TracePilot can time it and capture any errors that are thrown.

string

required

The spanId of the parent span. Tool calls must always be linked to a parent — typically the spanId returned by the wrapOpenAI call that decided to invoke this tool.

number

required

An integer that controls where this span appears among its siblings in the execution tree. Use sequential integers (1, 2, 3…) to preserve the correct order in the dashboard.

boolean

When true, the dashboard marks this span with a ⚠ Destructive badge. Use this flag for any operation that modifies external state — sending emails, writing to a database, publishing messages, calling payment APIs, and so on. Defaults to false.

Return value

The value resolved by your call function, returned unchanged. The generic type T is inferred from your function’s return type.

string

The ID of the span created for this tool call. You can pass this as parentSpanId to downstream wrapOpenAI or wrapToolCall calls to nest them further.

Examples

import { TracePilot } from 'tracepilot-sdk';

const tp = new TracePilot(process.env.TRACEPILOT_API_KEY!);

async function runAgent() {
  await tp.startTrace('notification-agent');

  // ... prior LLM call returns planSpanId

  const { result, spanId } = await tp.wrapToolCall(
    'send-email',
    () => sendEmail(to, subject, body),
    planSpanId,
    2,
    true  // marks this span with a ⚠ Destructive badge
  );

  console.log('Email sent:', result.messageId);
}

import { TracePilot } from 'tracepilot-sdk';

const tp = new TracePilot(process.env.TRACEPILOT_API_KEY!);

async function runAgent(query: string) {
  await tp.startTrace('research-agent');

  const messages = [{ role: 'user', content: query }];

  const { result: plan, spanId: planSpanId } = await tp.wrapOpenAI(
    () => openai.chat.completions.create({ model: 'gpt-4o', messages }),
    messages,
    undefined,
    1
  );

  // Read-only search — no isDestructive flag needed
  const { result: searchResult, spanId: searchSpanId } = await tp.wrapToolCall(
    'web-search',
    () => webSearch(plan.choices[0].message.content),
    planSpanId,
    2
  );

  return searchResult;
}

Set isDestructive: true for any tool that modifies external state. This includes — but is not limited to — sending emails or SMS messages, writing or deleting database records, publishing to message queues, making payment API calls, posting to social media, and modifying files on disk. The badge serves as a safety signal during Fork & Rerun sessions, letting you quickly identify which spans had real-world side effects.

​Signature

​Parameters

​Return value

​Examples

Signature

Parameters

Return value

Examples