A Burp Suite extension that pseudonymizes HTTP requests and responses so you can paste them into an LLM without leaking client data, and restores the LLM's answer back into real values so it can go straight into Repeater.
Right-click any request or response → Send to Anonymizer.
POST /api/v2/login HTTP/1.1 POST /api/v2/login HTTP/1.1
Host: portal.acmecorp.com Host: portal.target1.example
Cookie: session=4f3c…9a; lang=en Cookie: session=[COOKIE_1]; lang=[COOKIE_2]
Authorization: Bearer eyJhbGciOiJIUzI1… Authorization: Bearer eyJhbGciOiJIUzI1…REDACTED_SIGNATURE_1
Content-Type: application/json Content-Type: application/json
Content-Length: 83 Content-Length: 86
{"email":"[email protected]", {"email":"[email protected]",
"password":"Winter2024!", "password":"[PASSWORD_1]",
"company":"Acme Corp"} "company":"[CLIENT_1]"}
Two properties make this more than redaction:
- Consistent. The same host, cookie, token or email always becomes the same placeholder within
a session, so the model can still reason about the traffic:
[email protected]logs in, gets[COOKIE_3],[COOKIE_3]is replayed againstapi.target1.example. - Reversible. Every replacement is kept in a session map. Paste the LLM's reply into the Restore pane → Restore → Send to Repeater.
The extension is passive: it never sends traffic. "Send to Repeater" opens a tab; you send. It also never talks to an AI provider: it prepares text for you to paste into whichever model you use, and reads back whatever you paste in. Nothing leaves Burp except through your clipboard.
Hosts and domains (generic subdomains such as api, www, auth are kept; well-known CDNs and
third parties are kept), public IPs (private ranges optional), emails, Authorization headers,
cookies, JWTs (re-issued as synthetic tokens whose claims are anonymized by the same rules so alg,
exp, role stay readable), sensitive headers by name, sensitive JSON/form/XML/cookie fields by
key (password, token, ssn, card…), known secret formats (AWS, GitHub, Stripe, Slack, private
keys…), UUIDs, long hex strings, Luhn-valid card numbers, SSNs, phone numbers, opaque high-entropy
blobs, and your own client terms (company name, codenames).
Placeholders keep the value's shape where it matters: hosts stay valid hostnames, emails valid emails, UUIDs valid v4 UUIDs, JWTs three base64url segments. The request the LLM crafts still parses.
- In Proxy, Repeater, Logger, the Site map or Intruder, right-click a message → Send to Anonymizer. Select several first to send them all.
- In the Anonymizer tab, review the Anonymized pane. Blue is a placeholder; yellow is "check this", a value a disabled detector would have caught. Toggle detectors or add client terms (select text in any editor → Add selection as client term); the output re-renders live.
- Copy, or select several items → Copy all as Markdown, and paste into the LLM.
- Paste the LLM's reply into Restore → Restore. Unknown placeholders are left in place and listed (restore never invents a value); lenient matches are highlighted; a JWT whose claims the LLM edited is flagged rather than guessed. Then Send to Repeater.
The session map lives with the Burp project. Temporary projects (Burp Community) lose it on close: Export it first, Import it later. Clearing the map restarts numbering; nothing else does. Detector toggles follow you across projects; client terms stay with the project.
RequestRedactorX copies a request with
headers and chosen parameters blanked out ([REDACTED]). That is the right tool for sharing a
request in a report. Anonymizer solves a different problem: working with an LLM on live traffic.
| RequestRedactorX | Anonymizer | |
|---|---|---|
| Output | one-way redaction, same placeholder for every value | pseudonyms: a distinct, stable placeholder per value |
| Same cookie in two requests | both [REDACTED] |
both [COOKIE_3], so the model can follow the session |
| Hosts, emails, UUIDs, JWTs | blanked | shape-preserving, so the model's crafted request still parses |
| Values found by shape | no, only by configured name | yes: secrets, JWTs, card numbers, hex, high-entropy strings, your client terms |
| Responses | no | yes, including Set-Cookie, JSON, HTML inputs |
| Getting the answer back | manual | paste → Restore → Send to Repeater, with unknown or modified placeholders flagged |
Requires JDK 21 (Gradle provisions one if missing) and the Montoya API (compile-only; never bundled).
./gradlew shadowJar # → build/libs/anonymizer-<version>-all.jar
./gradlew test # engine fixtures + unit tests
Load the JAR in Burp via Extensions → Installed → Add. After rebuilding, use Remove and Add; Burp's auto-reload does not pick up a rebuilt JAR.
The engine (com.anonymizer.engine) is pure Kotlin with no Burp dependency; it is text in, text out
and its behaviour is pinned by a fixture corpus (src/test/resources/fixtures). Every
anonymize/restore runs on one engine thread, so numbering is deterministic. JSON is rewritten in
place and line endings are kept per line, so key order, whitespace and malformed bodies survive
byte for byte.
See docs/PLAN.md for the full design and decision log.
GPLv3. The Montoya API is compiled against under the Burp Suite licence and is not bundled.
