Instruments captures the evidence.
RocketTrace shows you where to investigate first.
Turn dense Instruments recordings into ranked, measured leads — investigate the highest-impact evidence first, then add another run to see what the next recording measured.
Free download · First three analyses with findings included · Requires macOS 26 and Xcode 27 or newer
806 traces analyzed with RocketTrace.
7,179 findings surfaced with RocketTrace.
668 agent prompts copied with RocketTrace.
From a wall of graphs to ranked evidence.
A ten-second trace produces millions of samples across dozens of lanes. RocketTrace reads all of it, keeps Instruments as the source of truth, and surfaces where your investigation deserves to start.
One real recording, opened in both apps. Findings describe measured trace state, not automatic code conclusions.
Same recording. Same prompt. Measured with and without RocketTrace.
We gave two AI models one Instruments trace and asked for what could be improved, with and without RocketTrace, then scored every answer against evidence in the recording.
- Lower token usage
- 5.6x
- Issues found (of 20)
- 19 vs 12
- Faster to findings
- 2x
Claude Opus 5.5
| Median of 3 runs | Without | With RocketTrace |
|---|---|---|
| Issues found | 12 of 20 | 19 of 20 |
| False positives | 1 | 0 |
| Tokens | 1766k | 312k |
| Elapsed | 9m 54s | 4m 45s |
GPT-5.6 Sol
| Median of 3 runs | Without | With RocketTrace |
|---|---|---|
| Issues found | 6 of 20 | 14 of 20 |
| False positives | 0 | 0 |
| Tokens | 2072k | 322k |
| Elapsed | 11m 45s | 5m 07s |
With RocketTrace, both models used fewer tokens and finished sooner, and found more of the planted issues on this trace. One trace, one app, two models, three runs each. Ratios compare within each model; the smaller one is shown. Savings during later fixes are not measured.
From early users
Ranked evidence, then a grounded fix.
“I'd spent the previous week on scrolling hitches without RocketTrace, and agents needed a lot of guidance to make confirmed progress. With RocketTrace I dumped the prompts into Codex and it worked through each item cleanly—after one session I couldn't notice any hitches, even on a debug build.”
D. Beck“I used RocketTrace to track down frames dropping at launch. It ranked the main-thread stalls and pointed me right to the area I needed to fix, and the AI prompts saved me a lot of time—after the fixes, a new recording showed the biggest stall was gone.”
J. Franco- iOS Developer · 2× Swift Student Challenge
- Building Haven & Lume
“I imported Instruments traces into RocketTrace and used its agent prompts to fix several concurrency issues. After resolving them and profiling again, the UI is much smoother where it identified issues, and viewing extracted JPGs is now more or less instant.”
T. Evensen- Developer of RawCull and RawCullFB
Portable investigation evidence
Built for how code gets written now.
Coding agents make shipping features faster — and generated code can be correct yet slow in ways reading it will never reveal. The more code you generate, the more measuring real runtime behavior matters.
RocketTrace closes that loop: it turns the analysis into a focused prompt with measured impact, representative frames, and source-search terms. Paste it into any coding agent to begin a grounded investigation without handing the agent a raw Instruments recording. RocketTrace also installs a rockettrace CLI and Agent Skill from Settings so agents can record and read results directly.
- Measured durations and event counts, not vibes
- Representative app-code backtraces and search terms
- Ground rules that keep measured leads from becoming assumed fixes
- Optional CLI and Agent Skill for structured record-and-inspect workflows
Investigate sustained main-thread CPU work
EVIDENCE
1.82 s of app-attributed work · 8.6% of recording
REPRESENTATIVE BACKTRACE
→ GalleryViewModel.thumbnail(for:) [MyApp]
GUIDANCE
Inspect the symbol and its call sites. Treat this direction as a hypothesis; a no-change conclusion is acceptable when the work is necessary.
Analysis stays on your Mac.
Recording, parsing, and analysis all run locally using the Xcode you select. No RocketTrace cloud account, no automatic uploads, no cloud processing — trace content leaves your Mac only when you explicitly attach it to a feedback report.
- Local analysis
- Trace parsing runs on your Mac with your selected Xcode's tooling.
- No automatic uploads
- Trace content, app identity, and prompts are excluded from analytics. Feedback attachments are opt-in.
- Your agent, your choice
- Prompts are plain text. Paste them into whichever coding agent you trust.
How it works
Profile. Investigate. Record Again.
Record or import a trace
Record macOS apps, iOS or iPadOS apps on connected physical devices, and iOS Simulator builds (Xcode 27 or newer) directly, or import a compatible Instruments trace from macOS, iOS, and iPadOS apps recorded with Xcode 26 or newer.
Watch findings stream in
Results appear while analysis runs. Every finding carries its measured impact — durations, counts, backtraces — and RocketTrace says so when a lane has nothing that deserves attention.
Investigate, then add a New Run
Agent Prompt packages the evidence and ground rules for your coding agent — or use the rockettrace CLI and Agent Skill from Settings. After a candidate change, choose New Run, select a newer matching build, and retain what the next recording measured alongside the earlier run.
See the current product guides →Read profiling workflows on the blog →
What RocketTrace detects
42+ performance problems, one ranked list.
One recording is analyzed across six evidence categories. Every finding names the problem, carries its measured impact, and keeps the backtrace that makes it actionable.
- 7 detectionsSwiftUI updates & redrawsViews redraw far more often, or more expensively, than expected.
- 8 detectionsSwift Concurrency mistakesasync/await code that is correct but slower than it should be.
- 2 detectionsAnimation hitches & scrollingScrolling stutters and animations skip frames.
- 9 detectionsHangs & unresponsivenessThe app freezes, taps are ignored, or the spinner appears.
- 15 detectionsCPU hotspots & wasteful workThe app feels slow, runs hot, or drains battery.
- 1 detectionApp launchThe app takes too long from tap to first usable frame.
Your app’s own context
Your signposts connect the trace to the work that matters.
RocketTrace pairs app-authored OSSignposter begin and end points into measured intervals. See every occurrence on the recording timeline, then open the SwiftUI, Swift Concurrency, CPU, hang, or hitch evidence that overlapped it.
Duration alone does not make an operation slow. Signposts provide the name and timing; ranked findings across SwiftUI, Concurrency, CPU, hangs, and hitches provide the evidence.
See how RocketTrace reads Signposts →Add Signposts with OSSignposter →
Direct recording
Record without leaving RocketTrace.
The optional Recorder Companion is a free, notarized helper for RocketTrace users who want to record Mac apps, connected physical iPhone and iPad targets, or iOS Simulator builds directly — pick a build, press record, and investigate the ranked evidence when you stop. Importing and analyzing existing traces works without it.
- Free and notarized
- A separate download, signed and notarized by Apple, installed in under a minute.
- One recording pipeline
- The companion runs the same Instruments tooling you already trust, using the Xcode you select.
- Updates itself
- The companion keeps itself current independently of App Store releases, so recording rarely needs your attention.
Before you trace.
Does RocketTrace replace Instruments?
No. RocketTrace uses the Instruments tools in Xcode and turns their recorded evidence into a focused, ranked investigation workflow.
Does it change my code automatically?
No. A finding is a measured lead, not a required fix. RocketTrace gives you evidence and an investigation prompt; you or your coding agent still inspect the code and decide what should change.
Which app platforms does RocketTrace support?
Direct recording currently supports macOS apps, iOS or iPadOS apps on connected physical devices, and iOS Simulator builds (Xcode 27 or newer). RocketTrace can import and analyze compatible Instruments traces from macOS, iOS, and iPadOS apps recorded with Xcode 26 or newer. The RocketTrace app itself requires macOS 26 and an active Xcode 27 or newer.
How does in-app recording work?
Recording is powered by the optional Recorder Companion, a small notarized helper that starts Instruments recordings for Mac apps, physical devices, and iOS Simulators. Importing and analyzing existing trace files works without it. You can download the companion from this site at any time.
Does trace content leave my Mac?
Recording and analysis stay on your Mac, and trace files, symbols, stack frames, source paths, app names, and prompts are excluded from RocketTrace analytics. Files leave your Mac only when you explicitly attach them to a Report Feedback submission.
Who builds RocketTrace
Fifteen years of profiling, built into a tool.
RocketTrace is built by Antoine van der Lee — shipping Apple-platform apps since 2009 and profiling them with Instruments since 2011. Every detection encodes a pattern learned from real traces, so you start from experience instead of an empty timeline.
Antoine also writes the SwiftLee blog for hundreds of thousands of Apple developers and builds RocketSim, used by 25,000+ developers and 250+ teams. His other tools include MCP Beast and Video Compressor.
Spend less time reading the trace.
Record a Mac app, a connected physical iPhone or iPad, or an iOS Simulator directly, or import a compatible macOS, iOS, or iPadOS trace recorded with Xcode 26 or newer. RocketTrace requires macOS 26 and an active Xcode 27+ toolchain.
Free download · First three analyses with findings included
Explore the blog →“I'd spent the previous week on scrolling hitches without RocketTrace, and agents needed a lot of guidance to make confirmed progress. With RocketTrace I dumped the prompts into Codex and it worked through each item cleanly—after one session I couldn't notice any hitches, even on a debug build.”
