🦊 YouSure Vision

Detects secrets using embeddings—no regex maintenance

Verifies credentials actually work—zero false positives

Correlates against breach databases—real-world context

Indexes all public code—instant breach cross-reference

Protects AI assistants—safety API for Copilot, Claude, etc.

The Problem We Solve 🦊

Secrets leak. We find them before attackers do.

💀

Developer Commits Secret

const API_KEY = "sk_live_abc123..."

→

🌐

Pushed to GitHub

git push origin main

→

👁️

Scraped by Bots

15 seconds to detection

→

🔥

Account Compromised

$47K AWS bill overnight

🦊 With YouSure

🔍

Detect

Embeddings catch secrets regex misses

→

✅

Verify

Actually test if credentials work

→

🗃️

Correlate

Check against 847M breach records

→

🛡️

Protect

Alert in seconds, not hours

847M+

Breach Records

<100ms

Detection Time

False Positives

∞

Pattern Coverage

The Five Layers 🦊

A comprehensive stack evolving from pattern detection to AI-powered credential intelligence

L5 AI Code Safety API

API for AI assistants to check code before suggesting. "Don't let Copilot suggest leaked secrets"

L4 Global Index (Shadow GitHub)

Pre-index all public repos with hashes/embeddings. When breach happens → instant cross-reference

L3 Breach Database Integration

Check against HIBP, breach collections. "This password appeared in 847 breaches"

L2 Active Verification

Actually try credentials in sandbox. Zero false positives—every alert is verified

L1 Embedding-Based Detection (OOTW)

Replace 58 regex patterns with learned similarity. Detects NEW formats without pattern updates

L0 Current: Pattern + Entropy + Intent

140 tests passing, 82.9/100 truth-manifesto. Production-ready foundation

L1 Embedding-Based Detection

Before (regex)

// Must add new pattern for sk_prod_
/sk_live_[a-zA-Z0-9]{24}/
/sk_test_[a-zA-Z0-9]{24}/
// Manual maintenance forever

After (embeddings)

train(["sk_live_abc123...", "sk_test_xyz..."]);
// 5 examples

detect("sk_prod_newformat...");
// Works! Never seen this format.

✓ No regex maintenance

✓ Detects novel formats

✓ Learns from examples

✓ Generalizes across services

L2 Active Verification

Metric	Without Verification	With Verification
Alerts	50 alerts	3 verified alerts
False Positives	96%	0%
Team Response	Ignores alerts	Every alert is actionable

When a secret is verified working, an AI agent explores blast radius: What permissions? What data is accessible? What's the worst case exploit?

L3 Breach Database Integration

Cross-reference against HaveIBeenPwned, breach collections, and historical leak databases.

Password: Welcome123!
├── Found in: 847 breaches
├── First seen: LinkedIn 2012
├── Rank: #1,247 most common password
└── Risk: CRITICAL - rotate immediately

L4 Global Code Index

Continuously clone all public GitHub. Store hashes and embeddings (not actual secrets). When breach happens:

-- Instant query, pre-computed
SELECT repo, file, secret_hash
FROM global_index
WHERE secret_hash IN (SELECT hash FROM breach_data)
   OR embedding <-> breach_embedding < 0.1;

-- Results in milliseconds, not hours

🔒 SHA-256 hashes (one-way)

🔒 Embeddings (not reversible)

🔒 Bloom filters (probabilistic)

🔒 NO actual secrets stored

L5 AI Code Safety API

AI coding assistants trained on GitHub. GitHub contains leaked secrets. AI might suggest code WITH real leaked secrets. The solution:

POST /v1/check
{
  "code": "const AWS_KEY = 'AKIAIOSFODNN7EXAMPLE';",
  "context": "code_generation"
}

RESPONSE:
{
  "safe": false,
  "reason": "Exact match to AWS key leaked in 2023 breach",
  "suggested_fix": "const AWS_KEY = process.env.AWS_ACCESS_KEY_ID;"
}

Integration targets: GitHub Copilot • Claude (via MCP) • Cursor • CodeWhisperer • Replit • CodeSandbox • StackBlitz

What Makes Us Different 🦊

🧠

Embeddings, Not Regex

Others maintain hundreds of patterns. We learn from examples and detect secrets we've never seen before.

✅

We Actually Verify

Others flag maybes. We spin up sandboxes and test if credentials actually work. Zero false positives.

🤖

AI Safety API

No one else does this. We stop AI assistants from suggesting leaked secrets in real-time via MCP.

Feature	GitGuardian	TruffleHog	YouSure 🦊
Pattern detection	✓	✓	✓
Entropy detection	✓	✓	✓
Embedding-based	✗	✗	✓
Active verification	Partial	--verify flag	Full sandbox
Breach correlation	Enterprise only	✗	✓
Global index	✓	✗	✓
AI safety API / MCP	✗	✗	✓

Business Model 🦊

🏢

Enterprise Contracts

~$100K/year

Companies pay for continuous scanning, verified alerts, and breach correlation across their entire codebase. White-glove security.

🤖

AI Safety API / MCP

Per-call pricing

AI coding assistants (Copilot, Claude, Cursor) call our API before suggesting code. We check for leaked secrets, breach patterns, vulnerable code. MCP integration for Claude.

🔬

Sandbox Testing

Freemium → Paid

We scan public GitHub repos in sandboxes, verify if secrets actually work, and notify repo owners. They sign up to see details and get future alerts.

🗄️

Continuous Monitoring

Subscription

We store findings and continuously monitor. When new breaches drop, we cross-reference and notify affected repos instantly. Protection that gets smarter over time.

The Flywheel

Scan GitHub → Find secrets → Notify devs → They sign up → Store & monitor → Future alerts ↺

Implementation Roadmap

Phase 1

Foundation (Current)

Pattern + Entropy + Intent 140 tests passing 82.9/100 truth-manifesto Correlation boost

Phase 2

Embedding Detection

Feature extraction Train on secret corpus Benchmark vs regex Integrate as layer

Phase 3

Active Verification

E2B sandbox Service verifiers Blast radius agent Rate limiting

Phase 4

Breach Integration

HIBP API Hash database Cross-reference Real-time alerts

Phase 5

Global Index

GitHub ingestion Hash/embedding storage Query infrastructure Incremental updates

Phase 6

AI Safety API

REST API design MCP tool for Claude Copilot/Cursor partnerships Rate limiting & pricing

🦊 YouSure

The Problem We Solve 🦊

The Five Layers 🦊

L1 Embedding-Based Detection

L2 Active Verification

L3 Breach Database Integration

L4 Global Code Index

L5 AI Code Safety API

Technical Architecture

What Makes Us Different 🦊

Business Model 🦊

Implementation Roadmap

Foundation (Current)

Embedding Detection

Active Verification

Breach Integration

Global Index

AI Safety API

Success Metrics