I Compared My Free Vanilla JS CBT Toolkit Against 5 Paid Mental Health Apps — Here Is the Architecture Difference

The toolkit: 36 free CBT tools, vanilla JS, zero signup, zero backend, zero dependencies. Live at alexcoledev.github.io/cbt-toolkit.

I built it because therapy apps cost $50-200/year and upload your thoughts to their servers. Here is how the architecture actually compares.

The Contenders

Tool	Type	Cost	Where your data lives
cbt-toolkit (mine)	36 vanilla JS tools	$0	Your browser (localStorage)
Woebot	AI chatbot	$35/mo	Their servers
Wysa	AI chatbot	$20/mo	Their servers
Moodnotes	Thought record app	$4 + IAP	iCloud (if synced)
Sanvello	CBT + meditation	$9/mo	Their servers
CBT-i Coach	Insomnia CBT	$0 (VA)	Their servers

Architecture Difference 1: Where Your Thoughts Live

This is the one that matters most to me.

Paid apps: You type "I feel like everyone is judging me" → it goes to their server → they run ML on it → they send back a response. Your raw thoughts are stored on someone else's infrastructure.

cbt-toolkit: You type "I feel like everyone is judging me" → the JavaScript runs in YOUR browser → localStorage.setItem() → it stays on your device. No network request. No server. No database.

// The entire "database" — 40 lines
const DB = {
  save(entry) {
    const records = JSON.parse(localStorage.getItem('cbt_records') || '[]');
    records.push({ ...entry, id: Date.now(), timestamp: new Date().toISOString() });
    localStorage.setItem('cbt_records', JSON.stringify(records));
  },
  getAll() {
    return JSON.parse(localStorage.getItem('cbt_records') || '[]');
  }
};

Your most vulnerable thoughts — the ones you write at 2am when you are anxious — never leave your machine. There is no API endpoint. There is no server log. There is no data breach surface.

Architecture Difference 2: Rule-Based vs ML

Paid apps (Woebot, Wysa): Use LLMs/NLP to generate responses. The advantage: personalized, conversational. The cost: non-deterministic (same input → different output), requires server compute, your data trains their models.

cbt-toolkit: Uses keyword-pattern matching against CBT-defined categories. The advantage: deterministic, zero-latency, zero-cost, explainable. The tradeoff: no conversational flexibility.

// The distortion detector — pattern matching, not ML
const DISTORTIONS = [
  { name: 'all-or-nothing', keywords: ['always', 'never', 'completely', 'total failure'] },
  { name: 'catastrophizing', keywords: ['what if', 'end', 'disaster', 'heart attack'] }
];

function detectDistortions(text) {
  return DISTORTIONS
    .filter(d => d.keywords.some(kw => text.toLowerCase().includes(kw)))
    .map(d => ({ distortion: d.name, reframe: REFRAMES[d.name] }));
}

Same input → same output. Every time. For a clinical technique where consistency matters, this is a feature, not a bug.

Architecture Difference 3: Accessibility

Paid apps: Download from app store → create account → verify email → subscribe → now you can use it. 5 steps before you see your first tool.

cbt-toolkit: Open URL → use tool. That is it. No install. No account. No email. Works offline after first load (it is just HTML/JS/CSS). Works on any device with a browser.

The Honest Tradeoff

I am not going to pretend my toolkit is better than Woebot at everything. It is not.

Where paid apps win:

• Personalized conversational responses (my tools give the same reframe every time)
• Professional therapist integration (some apps connect you to real therapists)
• Progress coaching and reminders
• Medically-reviewed content pathways

Where cbt-toolkit wins:

• Cost: $0 vs $240-2400/year
• Privacy: thoughts never leave your device
• Speed: instant, no network round-trip
• Transparency: the detection logic is open-source and auditable
• Accessibility: no signup wall, works on any browser, works offline
• Determinism: same thought → same analysis (reproducible)

When To Use Which

• Use a paid app if you want conversational support, therapist integration, or structured coaching programs
• Use cbt-toolkit if you want free, private, instant self-guided CBT exercises — thought records, distortion detection, safety behavior tracking, habituation logging, core belief drilling

They are not mutually exclusive. I use my own toolkit for quick thought records and a paid app when I want someone to talk to.

The 36 Tools

The toolkit includes:

• Cognitive Distortion Detector (10 distortions, keyword-pattern matching)
• Core Belief Detector (downward arrow technique, 13 terminal beliefs)
• Safety Behavior Detector (9 CBT behavior categories)
• Habituation Pattern Detector (time-series trend analysis)
• Catastrophe Prediction Calibration Detector (disconfirmation rate tracking)
• CBT Thought Record (5-column classic format)
• Panic Diary (with reframe generation)
• OCD/ADHD/PTSD/Shame/Anger specific tools
• ...and 28 more

All vanilla JS. All client-side. All free. All at alexcoledev.github.io/cbt-toolkit.

Why I Built It This Way

I wanted CBT tools that:

1. Cost nothing (CBT should not be paywalled)
2. Keep my thoughts private (they are the most sensitive data I have)
3. Work instantly (no loading screens, no API latency)
4. Are auditable (I can read the detection logic and know exactly why it flagged something)

Vanilla JS + localStorage is the simplest architecture that satisfies all four. No framework. No backend. No database. No API. No signup. Just HTML, JS, and your browser.

---

The toolkit is open-source: github.com/alexcoledev/cbt-toolkit. 36 tools, 12 detectors, zero dependencies. Star it if it helps.