Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.trodo.ai/docs/llms.txt

Use this file to discover all available pages before exploring further.

On NextJS with @vercel/otel? You can skip the Trodo SDK install entirely and use the env-var-only OTLP path — see OpenTelemetry / OTLP. The Vercel AI metadata mapping (userId / sessionId / agentName plus custom keys) lands in the same dashboard fields as the SDK path. The SDK install below remains the right choice when you want wrapAgent boundaries, feedback, or trackMcp.

​
What auto-instruments

The Vercel AI SDK does not use module-level monkey-patching; it emits OTel spans only when you explicitly enable telemetry per-call. Every call inside a wrapAgent with telemetry enabled becomes a span.
CallSpan kindAuto-extracted
generateTextllmmodel, provider, tokens, prompt, completion
streamTextllmSame, accumulated
generateObject / streamObjectllmmodel, tokens, schema, parsed object
tool() + tool-call looptooltool name, input, output
embed / embedManyllmmodel, tokens
Provider field is whatever provider adapter you use (openai, anthropic, google, mistral, …).

​
Install

npm install ai @ai-sdk/openai
No OTel package needed — the ai SDK includes its own OTel integration.

​
Minimum example

import trodo from 'trodo-node';
import { generateText, tool } from 'ai';
import { openai } from '@ai-sdk/openai';
import { z } from 'zod';

trodo.init({ siteId: process.env.TRODO_SITE_ID });

await trodo.wrapAgent('ai-sdk-bot', async (run) => {
  const { text, toolCalls } = await generateText({
    model: openai('gpt-4o-mini'),
    prompt: 'What is the weather in SF?',
    tools: {
      get_weather: tool({
        description: 'Current weather in a city',
        parameters: z.object({ city: z.string() }),
        execute: async ({ city }) => `Sunny in ${city}, 72F`,
      }),
    },
    experimental_telemetry: { isEnabled: true },  // required
  });
  run.setOutput(text);
});
The experimental_telemetry: { isEnabled: true } flag is required — without it, no spans are emitted. The Trodo SDK already has an OTel tracer registered, so the AI SDK finds it and pipes spans through automatically.

​
Rich metadata

await generateText({
  model: openai('gpt-4o-mini'),
  prompt,
  experimental_telemetry: {
    isEnabled: true,
    functionId: 'summarize-support-ticket',
    metadata: { tenant: 'acme', plan: 'pro' },
  },
});
  • functionId → span name
  • metadata → merged into the span’s attributes, searchable in the dashboard.

​
Tool calls

Every tool({ execute }) becomes a tool span with tool_name = the key you used in the tools object. The waterfall shows the alternating pattern with the LLM-returned toolCalls flowing into your execute handler.

​
Streaming

const result = await streamText({
  model: openai('gpt-4o-mini'),
  prompt,
  experimental_telemetry: { isEnabled: true },
});

for await (const delta of result.textStream) process.stdout.write(delta);
await result.text; // await this so telemetry flushes
Make sure to await the final text / usage promise — otherwise the span closes before tokens are recorded.

​
Auto vs manual cheat-table

OperationAuto?Notes
generateTextyesRequires experimental_telemetry: { isEnabled: true }
streamTextyesSame; await final text for token extraction
generateObject / streamObjectyesParsed object goes on span.output
tool({ execute })yeskind='tool', name from object key
embed / embedManyyes—
Custom provider adapteryesAs long as it returns standard usage

​
Gotchas

  • If you forget experimental_telemetry, everything still works — except the span is just never emitted. Add a lint rule or wrap in a helper.
  • The AI SDK records prompt and completion by default — if these are sensitive, disable per-call with experimental_telemetry: { isEnabled: true, recordInputs: false, recordOutputs: false }.
  • Tool name collisions across multiple generateText calls in one run are fine — each produces its own span.