LLM observability
Inspect prompts, responses, execution flow, and cost for AI SDK applications.
AI Tools · Const
Weights & Biases
Weights & Biases observability for the AI SDK is provided through Weave. It uses OpenTelemetry to capture AI SDK traces, including prompts, responses, flow, and cost, for inspection in Weave.
How it works
Weights & Biases is supported through Weave, which receives AI SDK telemetry through an OpenTelemetry backend. Configure the Weave endpoint, API key, and project, then register the OpenTelemetry integration and enable telemetry on supported AI SDK calls.
Syntax
registerTelemetryIntegration(new OpenTelemetryIntegration());Key points
- The documented Weights & Biases integration is Weave, an observability toolkit for LLM applications.
- AI SDK traces can be inspected in Weave after the integration is configured.
- Trace data includes prompts, responses, execution flow, and cost information.
- Configuration uses Weave as an OpenTelemetry destination for AI SDK traces.
- Telemetry is enabled per supported AI SDK function call through its telemetry option.
Common mistakes
- A Weights & Biases account is required before traces can be sent to Weave.
- The project identifier must combine the team and project names in the documented slash-separated format.
Example: Register the Weave telemetry integration
import { OpenTelemetryIntegration } from '@ai-sdk/otel';
import { registerTelemetryIntegration } from 'ai';
registerTelemetryIntegration(new OpenTelemetryIntegration());Registers the AI SDK OpenTelemetry integration so supported AI SDK calls can emit traces to a configured Weights & Biases Weave backend.
When to use it
Centralized trace logging
Send AI SDK traces to a Weights & Biases project.
Evaluation and improvement
Analyze and improve LLM application behavior using Weave.
Related
WeaveOpenTelemetryIntegration@ai-sdk/otelregisterTelemetryIntegrationtelemetry