Capabilities
Attack-surface mapping Verified exploits Auto-fix pull requests
Product
Findings feed Reports & CVSS Skills library
Resources
How it works FAQ Contact us Sign in Get started
Autonomous pentest engine · self-hostable

Break it before they do.

Rokkhe is an autonomous pentester. It maps your whole attack surface, proves real exploits with working proof-of-concepts, and opens the fix as a pull request.

The loop

From recon to remediation — on its own.

Point Rokkhe at a repository, a URL, or a cloud target. It runs the whole engagement in an isolated sandbox and hands you results you can act on.

01 — Map

See the whole surface

Rokkhe enumerates what an adversary would: routes, parameters, auth flows, dependencies, and misconfigurations — across every repo you point it at.

02 — Prove

Exploits, not guesses

Every finding ships with a working proof-of-concept, run against a sandboxed copy of your app. No theoretical risk, no false positives carried forward.

03 — Fix

The patch, as a PR

Rokkhe drafts the remediation and opens it as a pull request — CVSS breakdown, evidence, and a ready-to-review diff attached.

Point Rokkhe at your stack — your logos here
your-companyanother-coa-third-teamdesign-partnerearly-accessfifth-logo your-companyanother-coa-third-teamdesign-partnerearly-accessfifth-logo
See it work

What Rokkhe finds, how it proves it, how it fixes it.

Everything in one feed
A live, ranked view of what's exploitable across your repos.
Open 4 In progress 2 Fixed 12
CRITRK-231SQL injection in invoice export2h
HIGHRK-230IDOR on webhook secrets endpoint5h
MEDRK-227Rate-limit bypass via header spoofing1d
MEDRK-224Stack trace exposed in error response2d
Proof, not guesses
Each finding comes with an exploit, run against a sandbox.
exploit.py
$ python3 exploit.py
POST /api/invoices/export
  filter=acct' OR '1'='1
200 OK · 42,113 invoices, 1,204 orgs
$ 
The fix arrives as a pull request
Rokkhe opens the patch; you review and merge. Status stays in sync with the finding.
Mergedfix: parameterize invoice export query
@@ app/api/invoices/export.py
- db.query(f"… WHERE filter = {input}")
+ db.query("… WHERE filter = %s", (input,))
All checks passed · 6 checks · #1288 · +3 −2
FAQ

Questions, answered.

Rokkhe is an autonomous pentester. Point it at a target and it maps the attack surface, writes and runs real exploits to confirm what's vulnerable, and opens the fix as a pull request — the whole engagement, without a human driving each step.

Scanners flag patterns. Rokkhe proves them: every finding ships with a working proof-of-concept run against a sandboxed copy of your app, so you triage what's actually exploitable instead of a wall of maybes — and it drafts the patch, too.

Rokkhe only tests the target you point it at, within the scope you set — so aim it at staging if you'd rather not exercise production. The agents and their tools run in an isolated Docker sandbox, and Rokkhe is self-hostable, so your code and findings never leave your infrastructure.

Git repositories, live URLs, and cloud targets. Rokkhe reasons about routes, parameters, auth flows, dependencies, and misconfigurations the way an adversary would.

Bring your own key. Rokkhe works with OpenAI, Anthropic, Google, and Z.ai (GLM) among others — you choose the model, and the spend stays on your account.

Yes — Rokkhe is self-hostable. Run it on your own infrastructure and extend it with custom skills. Get in touch about self-hosting →

Get in touch

More Questions about Rokkhe?

Tell us what you are trying to secure and we will come back to you.

Know what's exploitable
before an attacker does.

Point Rokkhe at your code and see what it finds.