Trails is a tool that analyzes LLM agent execution traces to help developers understand failures and improve their agents. It processes agent runs to surface actionable insights without requiring manual inspection of individual traces.

The product provides aggregate analysis showing patterns across all runs, including stats by issue, intent, domain, and site. It automatically detects, tags, and groups issues for easy filtering and prioritization. Users can view step-by-step replays with screenshots, browser state, and the agent's reasoning, with errors highlighted.

Trails works by having users upload their agent's execution logs, which it parses from browser-use format. The system then analyzes these traces to automatically detect, tag, and group issues while identifying patterns across multiple runs.

The main benefit is helping developers fix agent issues faster by focusing on what matters most. Users can filter by issue type, drill into specific traces, and implement improvements more efficiently.

Trails targets developers working with LLM agents who need to debug and improve their performance. It handles browser-use format traces out of the box.