Skip to content

README

Jump to bottom
John Williams edited this page Mar 9, 2026 · 1 revision

👁️ ARGUS by googleadsagent.ai

Algorithmic Authenticity Analysis Platform

VirusTotal for fake social media profiles — free, open source, Cloudflare-native

What It Does

ARGUS analyzes any social media profile or post across LinkedIn, Reddit, Instagram, X, and Facebook and returns a trust score (0–100) with full reasoning. Flagged profiles get a public landing page indexed by Google. All analysis is framed as algorithmic opinion — not a verdict. Profile owners can dispute at any time.

Every page that goes live has been reviewed and approved by a human (you).

Architecture

Browser Extension (Chrome MV3)
    ↓ real-time badge injection
Cloudflare Worker API
    ↓ D1 database + KV cache
Analysis Queue (CF Queues)
    ↓ async processing
Detection Pipeline
    ├── Image Engine    (Hive AI + C2PA + EXIF)
    ├── Text Engine     (GPTZero + stylometrics)
    ├── Behavioral      (account age, footprint, patterns)
    └── Network Graph   (coordination detection via D1)
    ↓
Admin Approval Dashboard
    ↓ human review — you approve/reject
Public Landing Pages (Next.js on CF Pages)
    ↓
Google Indexing API (auto-triggered on publish)
    ↓
Evidence Vault (R2 — immutable archive)

Cloudflare Services Used

Service Purpose Free Tier
D1 SQLite database 5GB + 25M reads/day
R2 Evidence vault storage 10GB + 10M ops/month
KV Score cache + rate limiting 100K reads/day
Workers API + pipeline orchestration 100K req/day
Queues Async analysis jobs 1M messages/month
Pages Next.js frontend Unlimited

Monthly cost at scale: ~$0–$5

Setup

1. Prerequisites

npm install -g wrangler
wrangler login

2. Create Cloudflare resources

# D1 database
wrangler d1 create argus-db

# R2 buckets
wrangler r2 bucket create argus-evidence-vault
wrangler r2 bucket create argus-screenshots

# KV namespaces
wrangler kv:namespace create SCORE_CACHE
wrangler kv:namespace create RATE_LIMITER
wrangler kv:namespace create SESSION

# Queues
wrangler queues create argus-analysis-queue
wrangler queues create argus-notify-queue

3. Update wrangler.toml

Paste the IDs from step 2 into wrangler.toml.

4. Initialize database

wrangler d1 execute argus-db --file=./schema/d1_schema.sql

5. Set secrets

wrangler secret put ADMIN_SECRET_KEY      # Your private admin key
wrangler secret put HIVE_API_KEY          # hive.ai free tier
wrangler secret put GPTZERO_API_KEY       # gptzero.me free tier
wrangler secret put SEARCH_CONSOLE_KEY    # Google Search Console
wrangler secret put EMAIL_API_KEY         # Resend or Cloudflare Email

6. Deploy Worker API

wrangler deploy workers/api/index.js

7. Deploy Pages (Next.js frontend)

cd pages
npm install
npm run build
wrangler pages deploy .next

8. Load browser extension

  • Open Chrome → chrome://extensions
  • Enable Developer Mode
  • Load Unpacked → select extension/ folder

Admin Dashboard

Access at: https://argus.googleadsagent.ai/admin

The admin dashboard is your approval queue. Nothing goes public without your sign-off.

Approval flow:

  1. Community submits a profile URL via /submit
  2. System runs full analysis pipeline
  3. You get an email notification
  4. You review the score, signals, and evidence in the dashboard
  5. One click to approve (publish + Google index) or reject
  6. Submitter gets notified of outcome if they provided email

Legal Architecture

Every page is protected by five elements:

  1. "Algorithmic analysis, not a verdict" — displayed on every page
  2. Full methodology disclosure — open source, publicly auditable
  3. Confidence + false positive rate shown — transparency in scoring
  4. Prominent dispute mechanism — above the fold, not buried
  5. Human editorial review — you approved every published page

Language used throughout: "flagged for review", "high risk", "scored X/100 based on these signals" — never "fake", "fraud", or "criminal".

Dispute / Opt-Out Process

Anyone can submit a dispute at /dispute. Evidence tiers:

Tier Evidence Outcome
Tier 1 Government ID + platform verification Auto-approve removal
Tier 2 Cross-platform presence, employer confirmation Human review
Tier 3 "I am real" with no evidence Dispute denied

Resolved disputes convert the page to a "dispute resolved" notice — retained for transparency, score removed from public display.

Repository Structure

argus/
├── workers/
│   ├── api/index.js          ← Main Cloudflare Worker API
│   ├── pipeline/analyze.js   ← Detection engine pipeline
│   └── lib/
│       ├── utils.js          ← ID generation, rate limiting
│       ├── evidence.js       ← R2 evidence archiving
│       ├── seo.js            ← Sitemap + Google indexing
│       └── notify.js         ← Email notifications
├── pages/
│   └── app/
│       ├── admin/page.js     ← Admin approval dashboard
│       ├── profiles/[slug]/  ← Public profile pages
│       ├── submit/page.js    ← Community submission form
│       └── dispute/page.js   ← Dispute submission form
├── extension/
│   ├── manifest.json         ← Chrome MV3 manifest
│   ├── content_script.js     ← Badge injection
│   └── background.js         ← Service worker
├── schema/
│   └── d1_schema.sql         ← Cloudflare D1 schema
├── wrangler.toml             ← Full CF bindings config
└── README.md

Detection Engines

Engine Tools Detects
Image Hive AI, C2PA SDK, ExifTool GAN faces, missing provenance, EXIF anomalies
Text GPTZero, stylometrics AI-generated text, agenda concentration
Behavioral Custom scoring Age/activity ratio, footprint, posting patterns
Network D1 graph queries Coordinated networks, flagged clusters

Score weights: Network (35%) > Behavioral (30%) > Image (20%) > Text (15%)

MIT License

Open source. Community maintained. PRs welcome.

github.com/itallstartedwithaidea/argus

Clone this wiki locally