Skip to main content

Your first investigation

Use a small authorized dump first. Keep the completed capture immutable during analysis and open your source project for navigation.

Follow the memory​

  1. In Overview, read the size model, total versus reachable heap, and counts. Heap bytes are modeled object sizes, not file length or analyzer RAM.
  2. In Histogram, sort by shallow size to find storage-heavy classes, or retained size to investigate ownership.
  3. In Dominator Tree, expand a large owner. Child rows load on demand; nested retained totals overlap and must not be added.
  4. Use Why alive?, Inspect, and Show referrers to check roots, fields, and incoming references.
  5. Use Source to investigate the matching class. An owner is not necessarily the code needing repair.

A large object or a suspect is a lead, not proof of a leak. Validate lifecycle intent and comparable snapshots.

Run a query​

In Query, run:

SELECT object_id, class_name, shallow_size, retained_size
FROM instances
ORDER BY retained_size DESC
LIMIT 20

Compare a result with the inspector. IDs belong to this dump, not stable cross-capture identity. See the HeapQL reference.

All eleven views​

The current UI has Overview, Histogram, Dominator Tree, Leak Suspects, Waste, Source, Query, Compare, Timeline, Monitor, and AI Chat. Object Inspector is a panel, not a twelfth tab.

Host capabilities differ in source lookup, settings, exports and approval. Use the guide for your chosen host.

AI, reports, recovery​

Read AI privacy before configuring a provider or approving source. Model responses and fixes need review and testing.

Copy Incident Report is local but can include paths and previews; scrub it before sharing. Exports do not upload automatically.

If a server exits, use Retry for that dump. A slow analysis is not an exited process: choose Continue Waiting or Cancel when offered. See recovery.