Open HAR files

Drop HAR files here

HAR 1.1 or 1.2 (.har), up to 200 MB per file and 10 files at once. You can open several captures and compare them.

Your HAR files are processed locally in your browser.

Your HAR files stay on your device

HAR files often contain passwords, tokens and personal data, so they are read inside this browser tab, in a background worker that has no network access. This page’s security policy blocks background network requests, and nothing is saved: closing or reloading the page discards everything. After the page has loaded, it also works offline.

What is a HAR file?

A HAR (HTTP Archive) file is a JSON record of the network requests a browser made while you used a page: every URL, method, status code, header, cookie and timing, and often the request and response bodies too. Support teams ask for HAR files because they show exactly what happened between the browser and the servers.

That completeness is also the risk. A HAR file can contain session cookies, access tokens, API keys, passwords typed into forms and personal data from API responses. Anyone who receives it may be able to use them.

What you can do with this tool

  • See what the page talks to: every request classified with its evidence, first and third parties, and an endpoint inventory with IDs normalised.
  • Check API health: REST payload shapes and GraphQL operations, including errors returned with HTTP 200.
  • Follow related requests: journeys grouped by page, navigation and timing, and links observed when a value from one response is sent in a later request.
  • Find problems: slow and failing requests, payload sizes, redirects, duplicates, caching, media segments that fail or arrive late.
  • Check security and privacy: tokens, cookies and personal data, cookie attributes, security headers, data sent to third parties; then download a sanitized copy.
  • Compare two captures with volatile values normalised, and see confirmed changes and potential regressions.
  • Export an HTML, JSON or Markdown report, the findings as JSON, and the endpoints as CSV or JSON.

How to export a HAR file

Chrome and Edge

  1. Open the developer tools: F12, or Ctrl+Shift+I (⌘+⌥+I on Mac), and choose the Network panel.
  2. Tick Preserve log. Tick Disable cache as well if you want to see a first visit.
  3. Reproduce the problem, then select the Export HAR button (a downward arrow) in the Network toolbar.

Recent versions of Chrome export a sanitized HAR by default, without cookies and authorization headers. Exporting with sensitive data has to be turned on in the developer tools settings.

Firefox

  1. Open the developer tools (F12) and choose Network.
  2. Tick Persist Logs in the settings menu (the cog) if the page navigates.
  3. Reproduce the problem, then right-click any request and choose Save All As HAR.

Safari

  1. Turn on the Develop menu (Safari ▸ Settings ▸ Advanced ▸ Show features for web developers).
  2. Choose Develop ▸ Show Web Inspector ▸ Network, and tick Preserve Log.
  3. Reproduce the problem, then select Export in the Network tab.

Reading the results

Timing phases

  • Blocked: time spent queued in the browser before the request could start, for example waiting for a free connection.
  • DNS, TCP connection and SSL/TLS: setting up a connection. They are absent when an existing connection was reused.
  • Request sent: sending the headers and body.
  • Waiting (TTFB): from the end of sending to the first byte of the response. It includes server processing and the network round trip, and any proxy or CDN in between.
  • Content download: receiving the response body.

A HAR file uses -1 for phases that do not apply or were not measured. The tool shows these as “not recorded” and never estimates them.

Findings

  • Every performance finding names its threshold. The defaults (1 s for a request, 600 ms waiting time, 1 MB response) are starting points: change them on the Performance tab.
  • Evidence, not verdicts. A long waiting time suggests where to look; it does not prove a backend problem. A slow third-party request does not show that the third party caused an issue.
  • Failures are grouped by endpoint and status, with possible causes. A 401 followed by a successful retry, or a missing favicon, is usually harmless.
  • Comparisons classify each difference as breaking behaviour observed in the captured traffic, a performance regression, an informational difference, or unable to determine.

Before you share a HAR file

  1. Open it here and look at the Security & privacy tab.
  2. Create a sanitized copy. Remove response bodies too if the recipient does not need them: they often contain personal data.
  3. Search the sanitized file for anything specific to you that patterns cannot recognise: names, account or order numbers, internal host names.
  4. Share the sanitized copy only, and delete the original when you no longer need it.
  5. If an unsanitized HAR file was already shared, sign out of the affected sessions and rotate the exposed tokens and keys.

What automated detection can miss

  • Secrets under unusual names, or encoded, encrypted or split across fields.
  • Binary bodies (images, protobuf, compressed data), which are not searched.
  • Personal data that has no recognisable format, such as names and addresses in free text.
  • Data inside WebSocket messages and other fields some browsers add (they are copied as they are, apart from other copies of values found elsewhere).

What it checks

Read and checked

  • HAR 1.1 and 1.2 from browsers, proxies and test tools, including Chrome’s extra fields (resource type, transfer size, cache, errors).
  • Structure: malformed JSON, missing or invalid fields, invalid or negative timings. Entries that cannot be read are listed with the reason, never silently dropped.
  • Large captures: up to 200 MB and 100,000 requests per file, analysed off the main thread.

Not available from a HAR file

  • Why a server was slow. A HAR file records the browser’s view only.
  • JavaScript execution, rendering and layout. Use the browser’s performance profiler for those.
  • Exact domain ownership. Third parties are grouped by registrable domain using a built-in list of common suffixes; add your own domains as first party.
  • Timings the browser did not record, for example for cached responses.

Questions, answered

Are my HAR files uploaded?

No. They are read inside your browser tab, in a background worker that cannot make network requests, and the page’s security policy blocks background requests. Nothing is stored: closing or reloading the page discards the files. You can disconnect from the internet after the page has loaded and it still works.

Is the sanitized file safe to share?

It is much safer, but no automated tool can promise that every secret is gone. Detection is based on header and field names and on the formats of common tokens, keys and identifiers. Review the preview, remove bodies you do not need, and check the file for anything specific to you before sharing it.

Why does a request show “not recorded” timings?

The HAR file says so: -1 or a missing value means the browser did not measure that phase, for example because a connection was reused or the response came from the cache. The tool shows the gap instead of guessing.

How large a file can it open?

Up to 200 MB and 100,000 requests per file, and 400 MB across all open files. Memory is the limit: a HAR file needs several times its size in memory once read. On phones and older computers, smaller files work best. Ticking “Disable cache” and leaving out large media while recording keeps captures small.

Can it tell what the user did, or whether a video played?

No. A HAR file records the requests the browser made, not clicks or intent, and not whether media was decoded or played. Journeys describe groups of related requests, and a media segment with status 200 only means it was delivered. Everything the tool infers says so, with a confidence level and the evidence.

Why does a GraphQL request with status 200 show errors?

GraphQL servers usually answer 200 and report failures in an errors array in the body. The tool reads that array for each operation, independently of the HTTP status. If the response body was not recorded, the outcome is shown as unknown, not as a success.

Why is a request classified as “unknown”?

Because none of the rules matched. Classification uses only what is in the capture (body shapes, content types, the type the browser recorded, words in paths and host names) and never guesses. Open the request to see what was looked at.

How are two captures matched in a comparison?

By method, host and path. By default, IDs in paths, hashed file names, query parameter values and common cache-busting parameters are ignored, so /orders/123?_=171 and /orders/456?_=172 count as the same request. Each rule can be turned off.

More free tools

See all free tools
  • Developers
  • OpenAPI

API Contract Drift Analyzer

Compare two versions of an OpenAPI 3.0, 3.1 or 3.2 spec and see which changes break existing clients: removed operations, new required fields, type, enum, response and security changes, each explained.

  • Developers
  • Free

MCP Configuration Validator

Paste your MCP server config and get the exact line of every mistake: JSON errors, wrong keys, bad args and env values, exposed secrets and client-specific problems. Works for Claude, VS Code, Cursor and Windsurf.

  • Photos & videos
  • Use responsibly

Metadata Cleaner

Remove GPS location, device details, edit history, and AI generation details such as prompts and Content Credentials from JPEG, PNG, WebP, MP4, MOV, WebM and MKV files. No quality loss.

Chasing a slow or failing integration?

AICEA designs and builds APIs, integrations and the monitoring and tests that keep them fast and reliable.

Talk to us