Introducing RocketTrace: Using Xcode Instruments with AI Agents
Article Summary
Antoine van der Lee just launched RocketTrace, a Mac app that makes AI agents 6x more efficient at finding iOS performance issues. The benchmark results are striking: agents found more bugs while using a fraction of the tokens.
RocketTrace analyzes Xcode Instruments traces locally and converts them into ranked, actionable findings that AI coding agents (Claude, Cursor, Codex) can actually use. Instead of agents burning through 2 million tokens trying to parse raw trace files, they get compact summaries with the evidence they need to fix performance problems.
Key Takeaways
- Agents found 19 of 20 issues with RocketTrace vs 6-12 without it
- Token usage dropped 5.7-6.4x while analysis time cut in half
- Detects 42+ performance problems across CPU, hangs, Swift Concurrency, and SwiftUI
- Two-thirds of beta testers copied findings directly to their AI agents
- Everything runs locally with CLI integration, no cloud uploads required
RocketTrace helps AI agents find more iOS performance issues while using 6x fewer tokens, turning Xcode Instruments into something your coding agent can actually understand and act on.
About This Article
AI agents investigating Xcode Instruments traces without RocketTrace burned through tokens at an alarming rate. A typical investigation used 1.8 to 2.1 million tokens over 10-12 minutes as agents repeatedly exported data and re-read context.
Antoine van der Lee created RocketTrace to analyze traces locally before agents see them. The tool ranks findings by actual impact and gives agents compact JSON summaries with supporting evidence through a CLI, so they don't have to parse raw instrument data themselves.
In beta testing, 77 out of 100+ users finished their analyses. 49 of them copied agent prompts into their own workflows, and 21 used the 'Fix with your agent' feature. The numbers show people actually want this kind of integration between performance work and AI.