Back to public tools
All tools

Analyze a HAR network log

Find slow requests and large downloads in an existing network log.

Read a HAR network log you already exported. Inspect request timings and sizes, compare a selected request with a baseline, then review sanitized outputs. This tool never records or intercepts browser traffic.

Read this before choosing a file

HAR files can contain authorization headers, cookies, private URLs, form values and complete response bodies. Files stay in this browser’s memory. Do not share the original log without your own review.

Choose an existing HAR, or try the fully fictional example. Files allow 5,000 requests, 200 pages, 250,000 JSON values, 40 levels and up to two million characters per text field. Nothing is silently truncated.

How to inspect an existing HAR

Read the private-data notice before choosing a UTF-8 HAR or JSON file. Analyze HAR reads the current log locally. It never records your browser, intercepts traffic or visits imported URLs. Try the fictional example without importing a real recording. Original fields remain in memory; they are not saved as a device draft or uploaded to a service.

Filter by hostname, resource type, exact status or recorded page. Start order shows relative request positions; the other views place slow or large requests first, with unknowns last. The waterfall shows recorded start-to-finish spans for the current 50 rows, with a full timing/size table alongside it. Unknown starts are unplotted, unknown elapsed durations are circles, and zero durations are ticks. Tiny positive bars have a one-pixel minimum for visibility. Use the numeric table for exact values.

Inspect a request to see the timing breakdown. Original-field reveal is local text, limited to 12,000 characters, with no imported HTML execution. An optional baseline HAR has its own Analyze action and selected request. Differences compare just the two selected records; positive elapsed difference means the current record took longer. They are not a statistical result, an automatic URL match or proof of the cause of a slowdown. Baseline content is excluded from shareable current-log exports.

Example: a slow image and a redirect

The synthetic log has four fictional requests: a document, a large image, a redirect and a revalidated stylesheet. Its browser-reported transfer sum is 41,360 bytes, encoded body sum 41,000 bytes and decoded content sum 45,000 bytes. Slowest first puts the 800 ms image ahead of the 120 ms document. The stylesheet records zero elapsed time and zero transferred bytes; those are values, not missing measurements.

Select the image, load the synthetic baseline and analyze it. Compared with baseline request 1 at 100 ms, the selected image differs by 700 ms. Filter to the image hostname or one status, then clear those filters to inspect the full list. Export always covers the whole analyzed current log, independent of these table filters; its field-policy counts refer to that whole log.

What the timing and byte fields mean

The optional browser extension _transferSize supplies reported transfer bytes. response.bodySize supplies encoded response-body bytes; response.content.size supplies decoded content bytes. A separate body-plus-header estimate is available only when both sizes are known. They are labeled separately rather than treated as interchangeable totals. Byte sums include only known measurements and show how many requests are unknown.

Missing, negative, invalid or out-of-range timing values remain unknown; zero remains zero. The recorded entry elapsed time controls the waterfall. SSL is included in connection time and is not added to the phase sum again. A phase sum can differ from elapsed time when fields are absent or recorder conventions differ. Cache metadata, reported browser cache/service-worker flags and 304 responses are indicators; zero transfer alone does not prove a cache hit. These field conventions can be inspected in Chrome’s HAR exporter.

Review and download sanitized reports

Build sanitized preview reconstructs a fixed allowlist rather than exporting the original object. All request/response headers, cookies, body text and parameters, URL login details, fragments, source comments, IPs, connection IDs and arbitrary source extensions are removed. Query values are replaced. Defaults also anonymize hostnames, paths and query names, and shift timestamps to a fixed epoch while retaining relative offsets.

You can explicitly retain reviewed hostnames/ports, pathnames, query names or absolute timestamps. Those fields may themselves identify users, private systems or tokens. Review the field-by-field policies and sanitized preview before enabling downloads. Changing a choice invalidates the old preview and review check. The preview can be truncated; inspect retained values and review a saved file before sharing. Redaction is a review aid, not a guarantee that every private detail is found. Performance patterns and request order can also be sensitive.

Sanitized JSON preserves unknown measurements as null. Standalone HTML contains escaped report text and tables. Redacted HAR omits body/header/cookie data and uses representative MIME category labels. Unknown mandatory start/time/content values use explicit placeholders with generated unknown flags; this tool restores those unknowns when reopening its own redacted HAR. Recorder-specific information is intentionally lost. Exported files are separate from the in-memory original.

Supported HAR 1.2/1.1-style logs allow up to 20 MiB, 5,000 requests, 200 pages, 250,000 JSON values and depth 40. Text fields are limited to two million characters, URLs to 32,768, and a recording to a 366-day span. Worker processing stops after three seconds; each sanitized export is limited to 20 MiB. Cancel stops the current worker; invalid input preserves the last analysis with an explicit stale-source notice. Clear discards the workspace, restores the privacy notice and creates no background service.

Stars and feedback

Loading…

Tool reference #226