A native macOS performance monitor and recorder. See what your Mac is doing now, then go back to the moment a slowdown or spike began. Compare CPU, memory, GPU, network, disk, battery, and sensor readings with the processes behind them.

Use it as a live menu bar readout, a quiet background recorder, or a window you open when needed. The menu bar and history recorder have separate switches.

Free and open source. No usage telemetry. Recorded samples stay on your Mac.

2.2 release notes · Changelog · Documentation

This update adds more GPU history, an opt-in Ask preview, process usage timelines, and saved time ranges. The app still supports Apple silicon Macs on macOS 15 or later.

• GPU and Neural Engine: Total, Reads, and Writes share a historical bandwidth chart, marked Preview because the values are estimates from macOS buckets. ANE Time and ANE Power are separate readings and charts. ANE Time needs macOS 27; ANE Power needs the approved Full Coverage helper. GPU Memory and GPU awake now have recorded detail charts.

• Ask About This Mac: on macOS 27, ask a plain question such as "Why is my Mac slow?" or tap an area like Memory or Battery. The app checks its own readings and history, then Apple's on-device model explains them in everyday language, with links to the matching charts. Nothing leaves your Mac, and the area summaries work even when Apple Intelligence is off. Answers can still be wrong, so the facts behind each one are a click away.

• AI agents: hand a harder question to Claude Code, Codex or another agent. Ask copies a prompt that explains the recorded history, or the one-line setup for the bundled read-only MCP server:

  claude mcp add --scope user mac-performance-monitor -- "/Applications/Mac Performance Monitor.app/Contents/MacOS/mpm" mcp

  Run mpm help for the command-line tool. The AI agents guide covers Claude Code, Codex, Claude Desktop, Cursor and VS Code, what agents can ask, and what leaves your Mac.

• Usage Timeline: right-click a process to see when the recorder observed it running. Optional Apple app and media activity needs Full Disk Access and a per-window opt-in. It is not verified foreground-use history.

• Saved ranges: history views start at 30 minutes and remember their own range. Explorer also keeps your chosen zoom span.

New history grows from new recordings; older logs stay gaps for fields they never stored. See the release notes for requirements and limits, and the release checklist for verification details.

All six Energy cards open larger charts and explanations. Charge, runtime, Mac power draw, and battery temperature have their own detail ranges. Health and Cycles show daily history over months and years, kept separate for each battery pack. Long-term history grows from new readings with recording on.

Runtime uses macOS estimates or recent steady use. A dashed forecast stays separate from real charge data. It assumes the same workload, not a fixed promise of how long the battery will last.

The Accessories panel shows the battery levels macOS reports for mice, keyboards, AirPods, and other devices. Optional low-battery alerts live in Settings > Alerts. Checks run at most once a minute; unknown values stay unknown.

Process detail and Explorer can show earlier, non-overlapping runs of the same program, with gaps at restarts. The Dashboard also shows uptime and the boot date.

Energy guide

Explorer replaces the Analytics start screen with a workspace for live and recorded data. Pick a time, add the signals you need, and compare up to eight processes, including ones that have exited.

Hover to move a shared cursor across the charts. Click to pin a time. Hold Command and scroll to zoom around the pointer; ordinary scrolling still moves the page. Use a grid, a list, or an expanded chart for a closer look.

The inspector shows source values, timestamps, known bounds, and stored machine rows. Export the visible data as CSV, or share process history in a trace file. Older data keeps its retained resolution; zooming does not invent detail. Hardware inventory is a current snapshot, separate from history.

Explorer guide

High swap usage alone is not a reason to warn you. The new rules look for continued growth or paging strain. They can report further worsening without first waiting for usage to fall below an old fixed limit.

Process-growth checks reject stale readings and growth that has settled. Modest findings stay as quiet observations, not a diagnosis of a memory leak. A separate fast-growth check can catch runaway use without waiting 20 minutes.

Click the menu bar alert badge for active issues and their evidence. Snooze an ordinary alert for an hour, or open it in Explorer at the relevant time. Related memory alerts share a notice, and only critical notices request sound.

Adaptive alert rules and limits

Existing alert choices and history carry forward. The old swap threshold is no longer used, but explicit process-memory budgets remain available. Alert state is local and separate from full history recording. New database fields preserve detail from new samples; they cannot restore missing older readings.

Back up the app's data before testing a downgrade. Review the upgrade notes and privacy policy before sharing trace files, reports, or screenshots.

Dashboard and Explorer charts show a clear average over the translucent recorded range. Short bursts stay visible, missing readings stay missing, and historical shapes remain stable as new data arrives.

Dashboard cards open larger detail views with values and explanations. The menu bar panels keep stable layouts as readings change. The Dock icon follows the window, with a setting to keep it visible.

• Processes: sort and filter the process table, inspect memory and CPU history, and check file descriptors, disk I/O, and Rosetta status. Right-click a row and choose Usage Timeline for sampled running history. Optional Apple app and media activity appears separately, with unverified device and foreground status. It requires Full Disk Access and a per-window opt-in; the app keeps these activity records in memory only.

• Groups: collect related apps and helpers into a group and track their combined footprint as a share of the Mac's memory.

• Energy: view battery health, charge, power flow, temperatures, fans, and the processes using the most energy.

• Network: follow download and upload rates, inspect adapters, and enable per-app traffic tracking to see which apps use the network.

• Disk: chart throughput, IOPS, and service time. Inspect drive health, volume space, and the processes doing the most I/O.

• Disk Map: scan a disk or folder, explore its space as a treemap, and find large or old files. Reveal items in Finder or open Quick Look.

• GPU: see device and per-process activity, power, memory, and thermal limits. The bandwidth preview adds history beside separate ANE time and power, and recorded GPU memory and awake time. Recognize AI runtimes such as Ollama, MLX, and LM Studio.

• Hardware: browse a searchable inventory of the chip, memory, displays, storage, and connected devices. Refresh on demand or save a report.

• History and diagnostics: choose recording detail and retention, find top consumers over time, and investigate processes with on-device diagnostics.

GPU, Ask, and Usage Timeline examples use the real native views with sample data. They illustrate the interface, not a benchmark or a verified AI diagnosis. Capture details are in the release checklist.

The current overview: pressure, memory, CPU, network, disk, and thermal trends.

A live process table with a detailed inspector for the selected process.

Sampled running history and optional Apple activity occupy separate lanes. This example uses sample data, not a person's activity records.

Current measured reports remain available with AI off. This example uses sample readings. Questions, explanations, and Siri sharing have separate controls.

Battery health, power flow, and accessory batteries. This overview is cropped to leave the battery serial number out of the image.

GPU memory and awake-time history, separate ANE cards, and the bandwidth Preview chart. The approximation caveat appears once beneath the chart. This current-source capture uses sample readings.

ANE Time and ANE Power have independent units and history. These are sample readings, not evidence that an app used a particular processor.

For the latest published build, download MacPerformanceMonitor.pkg from Releases and double-click it. Published packages are Developer ID signed and notarized by Apple. Sparkle handles app updates.

brew install --cask mac-performance-monitor

This installs the same signed, notarized pkg from the main homebrew-cask repository. Homebrew's version can lag a new release until its cask update lands. The app also keeps itself current through Sparkle. To include it in brew upgrade, pass --greedy.

git clone --branch main https://github.com/Zesty0wl/mac-performance-monitor.git
cd mac-performance-monitor
swift build
swift test
Scripts/run.sh

You need Apple silicon and macOS 15 (Sequoia) or later. Building the app bundle needs Xcode 27 for its App Intents metadata and the macOS 27 SDK. Core builds and tests use Swift 6. Bundling needs Xcode's xcstringstool too. The run script uses a signing identity when available, or falls back to ad-hoc signing. Ad-hoc builds cannot use the privileged helper. See CONTRIBUTING.md for signing options and test coverage.

No usage telemetry or analytics. Recorded performance data and alert evidence stay on your Mac. Exports leave it only when you choose to share them.

Update checks, model downloads, signed content downloads, and network tools make network requests. They do not upload your recorded performance history. Ask uses local models and clears its conversation when closed. If you hand an investigation to an AI agent such as Claude Code, the agent sends what it reads to its provider. Optional Siri and Shortcuts sharing follows Apple's processing rules and can pass results to other actions. See Security and privacy for these separate choices.

Contributions are welcome. See CONTRIBUTING.md and the Code of Conduct. Security reports go through SECURITY.md.

The app ships in English, Simplified Chinese, German, and French. German and French began as AI translations and await review by native speakers. New 2.0 strings in Simplified Chinese also include generated copy awaiting review. Translate or review in your browser on Crowdin, or edit one file and open a pull request. See TRANSLATING.md for the translation history and review process.

Released under the MIT License. Bundles GRDB.swift (MIT) and Sparkle (MIT). Local inference also uses MLX and llama.cpp, with their dependency licences in the app bundle. Model downloads include their own licence and attribution files.