skip to content

ai-native secret sweeps · defensive only

Find your leaked secrets before someone else does.

Paste code, drop a repo archive, or point at a public repo. Leaksweep finds the credentials you shipped by accident, explains the real risk in plain language, and drafts the exact fix for your stack.

samples
engine onlinefiles swept 0secrets masked 0shown in full 0illustrative

Sweep only code you own or code that is already public. No login. Source is processed in memory and discarded; secrets are masked before anything is returned.

live sweep · runs for real

This is the product. Run it.

POST /api/sweep · 60 KB · 10 per 10 min

src/config/storage.js
ready
1// src/config/storage.js
2const { S3Client, PutObjectCommand } = require("@aws-sdk/client-s3");
3
4// Docs show keys like AKIAIOSFODNN7EXAMPLE. Ours live below for now.
5// TODO(sam): move to env before launch
6const s3 = new S3Client({
7 region: "us-west-2",
8 credentials: {
9 accessKeyId: "AKIAT4GXQ7ZL2MNP5RWD",
10 aws_secret_access_key: "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",
11 },
12});
13
14async function uploadInvoice(id, body) {
15 return s3.send(
16 new PutObjectCommand({
17 Bucket: "acme-prod-invoices",
18 Key: `invoices/${id}.pdf`,
19 Body: body,
20 })
21 );
22}
23
24module.exports = { uploadInvoice };
25

standing by

// a config someone debugged at 1am.
// one real key pair, one decoy.

Run the sweep to see patterns find the candidates and the reasoning layer decide which one is a live credential.

how it works

Three passes. One of them needs a model.

Finding strings that look like keys is pattern work. Knowing which ones matter, and what to do about each, is not.

$ leaksweep sweep

01

Sweep

Paste code, upload an archive, or give a public repo URL. 29 detectors and entropy gates pull every credential-shaped candidate. Each one is masked the moment it is found.

[ok] 214 files, 18,402 lines
[ok] 7 candidates, masked at source

$ leaksweep reason

02

Understand the real risk

The reasoning layer reads masked context: the file, the code around it, what it sits next to. It separates a live key from a fixture, rates the blast radius, and says why.

[!!] 2 critical, 1 high
[--] 4 cut as noise, reasons attached

$ leaksweep fix

03

Rotate and prevent

Every finding ships with ordered steps for your stack: rotate at the issuer, move to env, purge history, read the audit log. Then a CI config so it does not happen twice.

[ok] rotation steps copied
[ok] re-sweep: 3 sealed, 0 remaining

The long version, with the prompts and the guardrails: how the AI works.

the console

Everything a finding needs to become a fix.

detection

Pattern plus reasoning.

Provider-specific patterns for AWS, GitHub, Stripe, Slack, Google, OpenAI, SendGrid, npm and more. Private key blocks. Connection strings. Entropy-gated generic assignments. Then a model reads the context and decides what is real.

candidate detector entropy verdict
AKIA••••••••••••5RWD aws_access_key_id 4.22 live risk · critical
wJ7r••••••••••••5yWn aws_secret_access_key 5.32 live risk · critical
AKIA••••••••••••MPLE aws_access_key_id 3.68 noise · docs example
c8f1••••••••••••28e4 generic_secret 3.98 likely real · medium

masked by design

sk_l••••••••••••YhJ9

First four, last four. Never the middle. Not on screen, not in storage, not in a prompt, not in a log.

false-positive cuts

4 of 7 cut

Placeholders, fixtures, and docs examples are dismissed with a stated reason. Still listed, never hidden, so you can overrule it.

remediation for your stack

Steps you can paste, in your idiom.

Django settings get os.environ. Next.js gets process.env and a note on the public prefix. Rotation names the exact console page.

git history purge guidance

$ git filter-repo --invert-paths --path .env
$ git push --force --all

Rotation first, always. Then the exact rewrite for that file, the force-push, and the re-clone note for your team.

ci prevention

A pre-commit hook and a pull request gate, with correct config for GitHub Actions and GitLab CI.

read the ci docs →

model-agnostic

A thin router picks a healthy provider and falls back down a chain. No model at all? The rules floor still returns correct fixes.

built ai-native

The model is the product, not a feature.

Risk judgement, noise cuts, and stack-specific fixes do not exist without a language model. We built around that from the first commit.

our stance

Defensive only.

  • >Secrets are never shown in full. Not to you, not to us, not to a model.
  • >Found something? Rotate it immediately. A masked report does not make a leaked key safe.
  • >We never test a credential against a live service, and we refuse sweeps aimed at code you do not control.
  • >The full position is on the security-first care page.

from developers

What it is like on the receiving end.

/** * Swept a four year old monorepo on a Friday. * It found a Stripe live key in a seed script and * told me which dashboard page rolls it. It also * waved off 11 fixture passwords with reasons I * agreed with. Rotated before lunch. */
@priya.raman · staff engineer, logistics startup
/** * Other scanners gave me 60 hits and no opinion. * This gave me 3 that mattered, the filter-repo * command for my exact .env path, and a CI file * that has blocked two commits since. */
@tomas.lindqvist · founder, two person SaaS

pricing

Priced like a tool, not a platform.

all prices USD · cancel any time

Free

$0USD / month

The full engine, three times a month.

  • +3 sweeps a month
  • +Paste up to 200 KB, archives up to 1 MB
  • +AI risk reasoning and false-positive cuts
  • +Masked findings with remediation steps
  • +Re-sweep after rotation, with delta
  • +Markdown report export

Dev

most picked

$19USD / month

For the developer who owns the fix.

  • +Unlimited sweeps
  • +Public repo URL sweeps (GitHub, GitLab)
  • +Archives up to 4 MB, paste up to 1 MB
  • +Git history purge plan per finding
  • +CI config generated for your stack
  • +JSON and Markdown report export

Team

$59USD / month

For a team that ships every day.

  • +Everything in Dev
  • +25 saved repos with one-key re-sweep
  • +5 seats on one workspace
  • +Priority lane on the model router
  • +Priority support, one business day
  • +Security review call on request

why now

Scrapers read every public commit within minutes.

A key pushed to a public repo is harvested long before a human reviews the pull request. The only thing that beats a scraper is finding it first, and knowing exactly what to do in the next ten minutes.

faq

Questions a security-minded developer asks.

The scanner sees the value for the length of one request, in memory, to mask it and compute a keyed fingerprint. After that it is gone. The response, the database, the logs, and the model prompt only ever contain the masked form: first four and last four characters. Short values show even less.

$ leaksweep init

Sweep your code before someone else does.

Three sweeps a month on the free tier. No card. Secrets masked from the first byte.