Redact a JSON file

It parses your JSON and walks the whole tree — objects, arrays, any nesting depth. Values under credential-shaped keys (password, secret, token, auth, api_key and relatives) are masked whole, even when the value looks harmless, and even when it is not a string: a numeric auth code becomes "[SECRET-1]", and an entire nested credentials object collapses to a single placeholder — the key survives, the shape of what was under it does not, and the copy says so rather than pretending otherwise. Every other string value is scanned by the same pattern engine that powers this site’s log file redactor: emails, IPv4/IPv6 addresses, JWTs, AWS keys, Bearer tokens, and key=value credentials embedded inside strings (a form-encoded body like password=…&grant_type=password gets the credential masked and the harmless grant_type left alone). Long random-looking strings that match neither pass are flagged for review with their JSON path — never silently changed.

Masked values become deterministic placeholder tokens, and the same value always becomes the same token: an API key that appears in config.api_key and again in a request header three levels deep shows up as the same [SECRET-1] in both places, so “these two fields held the same value” — often the whole diagnosis — survives redaction. The output is valid JSON at the indentation you choose (two spaces by default, four, or minified), keys and structure stay put, and redacting an already-redacted file changes nothing.

The web is full of excellent JSON formatters, validators, viewers, and converter toolkits — and none of them ship a redaction recipe. They will pretty-print your API response beautifully, secrets included. The gap is not formatting skill; it is curation: knowing that password, client_secret, and x-api-key are credential-shaped names, that a three-segment eyJ… string is a JWT, and that the same token must be masked consistently everywhere it appears. That curated knowledge is this page’s entire job — and doing it in a tab that provably cannot phone home is the other half, because an online tool you upload secrets to for cleaning is the problem wearing a different hat.

This site’s .env & config scrubber also handles JSON — flat, config-style JSON, which it edits line by line so everything outside the masked values stays byte-identical. This page is for the other kind: deeply nested JSON — API responses, exports, payloads — where line-based editing cannot follow the structure, at the honest price that the file is parsed and re-serialized rather than preserved byte for byte. For the other shape of structured data — a tabular CSV export with PII in columns — the CSV sample anonymizer is the sibling tool, masking whole columns instead of walking a tree.

Detection is pattern-based and incomplete by construction; any tool claiming otherwise is overpromising. The specific gaps: a secret under a bland key — endpoint, dsn, a vendor-specific name the heuristic doesn’t know — is only caught if the value matches a pattern or is long and random enough for the suspicious-string flagger (random strings under about 20 characters slip beneath its statistical floor). Prose-shaped passwords are invisible to it. Booleans and null are deliberately never masked, even under sensitive keys — they carry at most one bit — so "auth": true stays. And it can over-mask: a vault placeholder under a password key gets masked like any other value, because masking too much is the safer failure. Strict JSON only — JSONC, JSON5, comments, and trailing commas are rejected with a positioned error, never silently repaired. Re-serialization normalizes formatting: escape sequences become literal characters, number spellings are canonicalized (integers beyond 2^53 lose precision), and duplicate keys keep only their last value. The side-by-side preview highlights every change so the final human review takes minutes, not an afternoon of scrolling JSON.

Files are processed entirely in this tab’s memory with a 20 MB ceiling. Scanning runs on the page’s main thread, so a large file can pause this tab for a few seconds — nothing is being uploaded during that pause; there is no network activity at all, which you can verify in DevTools. Documents nested thousands of levels deep are refused with a clear error rather than crashing the tab. The page works offline once loaded, and nothing persists after the tab closes.