Weights & Biases Weave

Weave by Weights & Biases is an LLM observability toolkit that provides tracing, evaluation, and dataset management for AI applications. It integrates with the broader W&B experiment tracking ecosystem.

Weights & Biases Weave is an observability and continuous improvement platform designed specifically for production LLM agents. It provides end-to-end tracing, evaluation, and dataset management to help teams monitor agent behavior, surface failure modes, and prevent regressions. Weave organizes traces into sessions and turns, treating agents, tools, and sub-agents as first-class concepts, which enables deep analytics for multi-turn, multi-agent systems. The platform integrates with the broader Weights & Biases experiment tracking ecosystem and supports coding agents like Claude Code for autonomous self-improvement via W&B skills and an MCP server.