@agentlair/vercel-ai

AgentLair telemetry integration for the Vercel AI SDK. Captures tool calls, completions, errors, and token usage from your AI SDK generations and forwards them to AgentLair to build your agent's behavioral trust profile.

Zero runtime dependencies. Works with any AI SDK model provider.

Install

npm install @agentlair/vercel-ai
# or
bun add @agentlair/vercel-ai

Quick start

bun add @agentlair/vercel-ai ai @ai-sdk/openai

import { generateText } from "ai";
import { openai } from "@ai-sdk/openai";
import { createAgentLairExporter } from "@agentlair/vercel-ai";

const agentlair = createAgentLairExporter({
  apiKey: process.env.AGENTLAIR_API_KEY!, // al_live_...
  agentId: "my-agent",
});

const { text } = await generateText({
  model: openai("gpt-4o-mini"),
  prompt: "Summarise last week's sales",
  tools: { ... },
  experimental_telemetry: { isEnabled: true, tracer: agentlair },
});

await agentlair.shutdown(); // flush before process exit

Captures tool call timing, token usage, and session lifecycle — no prompt content leaves your process. Get your API key at agentlair.dev.

What gets captured

| AI SDK event | AgentLair event | Category | Data | |---|---|---|---| | onStart | generation_start | session | model, provider, temperature, tool count | | onToolCallStart | {toolName} | tool | tool name, call ID | | onToolCallFinish | {toolName}_complete | tool | duration, success/failure, error | | onStepFinish | step_complete | session | token usage, finish reason, tool call count | | onFinish | generation_complete | session | total token usage, finish reason |

Events are batched and flushed every 5 seconds (configurable). Buffer auto-flushes when it reaches 50 events.

Configuration

const agentlair = createAgentLairExporter({
  // Required
  apiKey: 'al_live_...',     // Your AgentLair API key
  agentId: 'my-agent',       // Agent identifier for event attribution

  // Optional
  baseUrl: 'https://agentlair.dev',  // API base URL
  flushInterval: 5000,               // Batch flush interval (ms)
  maxBufferSize: 50,                 // Force-flush threshold
  sessionId: 'custom-session-id',    // Group events by session
  timeout: 3000,                     // Request timeout (ms)
  captureInputs: false,              // Log input metadata (prompt length, message count)
  captureOutputs: false,             // Log output metadata (response length)
  onError: (err) => console.error(err),  // Error handler (swallowed by default)
});

Behavioral attestation

Events submitted through this adapter feed AgentLair's trust engine. Your agent's trust score reflects:

• Consistency of tool usage patterns over time
• Transparency through complete audit trails
• Behavioral adherence to declared capabilities

View your agent's trust profile at https://agentlair.dev/badge/{agentId}/score.json or embed a badge in your README.

License

MIT