memlab

Finding leaks in production at

Features

Everything you need to chase down memory

From automated end-to-end leak detection to low-level heap analysis — in the browser, in Node.js, and in CI.

Browser leak detection

Drive your app with the Puppeteer API. memlab diffs the heap snapshots, filters out the noise, and hands you the retainer trace of what actually leaked.

$ memlab run --scenario settings.js1 leak · 7.4MB

WindowGC root
.eventBusEventBusobject
.listenersArrayobject
.[0]handleThemeChange()closure
.dialogSettingsDialogcaptured variable
.containerDetached <div class="settings">1,205 detached nodes7.4MB

New

MCP server for AI agents

Hand heap snapshots to Claude Code, Cursor, or any MCP client and investigate memory in plain English.

agent · memlab mcp

The heap grows every time I open and close Settings. The snapshots are in ./snaps. What is leaking?

●memlab_load_snapshot(./snaps/*.heapsnapshot)

6 snapshots: a baseline + 5 open/close cycles

●memlab_leak_report()

Grows on 5 of 5 cycles:

SettingsDialog+1 / cycle

Detached HTMLDivElement+241 / cycle

Retained+7.4MB / cycle

●memlab_settle_check()

None of it comes back after idle + GC: a leak, not work in flight.

●memlab_dev_artifacts()

0 B retained only by DevTools or extensions.

●memlab_retainer_trace(SettingsDialog)

window

.eventBus.listeners[0]handleThemeChange()

.dialogSettingsDialog

.containerDetached <div> 7.4MB

SettingsDialog subscribes handleThemeChange to eventBus on mount and never unsubscribes. The listener closure keeps every closed dialog reachable from window. Return the unsubscribe from the effect:

useEffect(
  () => eventBus.on('theme', handleThemeChange),
  [],
);

Fixed. New snapshots are in ./snaps-fixed.

●memlab_verify_fix(./snaps → ./snaps-fixed)

Growth / cycle7.4MB → 0 B

Detached nodes1,205 → 0

✓ Leak fixed

Heap traversal API

An object-oriented API over snapshots from Chromium, Node.js, Electron.js, and Hermes — build your own leak detectors.

Memory CLI toolbox

Built-in analyses for duplicate strings, oversized objects, and unbound growth — optimization wins beyond leaks.

MemLens debugging

Visualize leaks and debug them interactively in the browser, without leaving the page you are testing.

Assertions in Node.js

Let unit tests snapshot their own heap and assert that objects are gone — catch regressions in CI, not in production.

How it works

Three steps to a leak trace

Write a scenario, run it, and read the reference chain that keeps your objects alive.

Define your test

Describe the interaction you want to check for leaks.

test.js

function url() {
  return 'https://www.google.com/maps';
}
async function action(page) {
  await page.click('text/Hotels');
}
async function back(page) {
  await page.click('[aria-label="Close"]');
}

module.exports = {action, back, url};

Run memlab in the CLI

One command drives the browser and reports leak traces.

zsh

$ memlab run --scenario test.js

# Analyze duplicated strings in the heap
$ memlab analyze string
# Check unbound object growth
$ memlab analyze unbound-object
# Get shapes with unbound growth
$ memlab analyze unbound-shape
# Discover more memory analyses
$ memlab analyze -h

Or use the API

Take snapshots and find leaks programmatically.

find-leaks.js

const {findLeaks, takeSnapshots} =
  require('@memlab/api');

async function test() {
  const scenario = {
    url: () => 'https://www.facebook.com',
  };
  const result = await takeSnapshots({
    scenario,
  });
  const leaks = findLeaks(result);
  // ...
}

Retainer traces

See exactly what is holding memory

memlab walks the reference chain from the GC root down to the leaked object, so you know which reference to break instead of guessing. Similar leaks are clustered, so 46 leaked objects become a handful of real bugs to fix.

Detached DOM and unmounted React fibers, flagged
Leaks clustered by retainer trace similarity
Retained sizes, so you fix the biggest first

Read the debugging guide →

memlab — heap diff

memlab comparing heap snapshots to surface a leak trace

Catch the leak
before your users do.

Read the docs Set up with your AI agent