BetterAsk

AI prompt autocorrect for ChatGPT, Claude, Perplexity, and Gemini.

BetterAsk is a Chrome extension that detects vague prompts before you send them, suggests clearer versions, learns from your editing patterns, and audits AI responses after they arrive.

---

Quick Start

cd betterask
npm install
npm run build

Then in Chrome:

1. Go to chrome://extensions
2. Enable Developer Mode (top right)
3. Click "Load Unpacked"
4. Select the betterask/dist/ folder

---

How It Works

1. You type a rough prompt in ChatGPT, Claude, Perplexity, or Gemini
2. BetterAsk detects if it's vague or missing key context
3. A green ✨ fix ask pill appears near the input
4. Click it — BetterAsk shows the original, an improved version, and why it helps
5. Click Use Better Ask to replace your text
6. BetterAsk saves the event locally for your dashboard

---

Features (Phase 1)

Feature	Status
Prompt detection (vague/weak)	✅
Local heuristic improvement (no API needed)	✅
API-enhanced improvement (OpenAI-compatible)	✅
Floating "fix ask" pill	✅
Suggestion card with edit	✅
Accept / Edit / Reject / Save Rule	✅
Local prompt event storage	✅
Saved rules (manual)	✅
Rules applied to suggestions	✅
Usage dashboard	✅
Response audit (local heuristic)	✅
Response audit (API-powered)	✅
Pattern learning (rule suggestions)	✅
Settings page	✅
Export / Import memory	✅
Delete all memory	✅

---

Supported Sites

• chatgpt.com
• claude.ai
• perplexity.ai
• gemini.google.com

---

Privacy

• No backend. All data stays in chrome.storage.local by default.
• No analytics. No third-party tracking.
• API keys are stored locally and never sent to BetterAsk servers.
• API mode is opt-in. When enabled, prompts are sent to your configured endpoint only.
• Use Export Memory (Settings → Privacy) to download your data as JSON.
• Use Delete All Memory to wipe all stored events and rules.

---

Known Limitations

• DOM selectors break. AI sites update their HTML regularly. Selectors are isolated in src/content/detector.ts and src/content/dom.ts so they're easy to update when that happens.
• Token/time savings are estimates. The dashboard uses characters / 4 to approximate tokens and 1.5 minutes per avoided follow-up as a configurable default. These are rough proxies, not measurements.
• Response auditing is not a guarantee of truth. The local audit uses heuristics. The API audit uses an LLM. Both can be wrong.
• A second model can still be wrong. Second opinion mode reduces hallucination risk but does not eliminate it.
• Contenteditable replacement. document.execCommand is deprecated but still works in most browsers. ProseMirror/Quill editors may need site-specific handling if the fallback breaks.
• Shadow DOM pointer-events. The floating UI uses a shadow root for CSS isolation. If clicks don't register on some sites, it may need tuning in src/content/index.tsx.

---

Project Structure

src/
  background/     — Service worker (background.js)
  content/        — Injected into AI sites
    detector.ts   — DOM detection (update selectors here when sites change)
    dom.ts        — Site-specific helpers
    floating-ui   — The "✨ fix ask" pill and suggestion card
    response-audit-ui — Post-response overlay
  popup/          — Extension popup
  options/        — Settings page
  dashboard/      — Local analytics dashboard
  shared/         — All business logic
    heuristics.ts — Local pattern-matching (works without API)
    improve.ts    — Orchestrates heuristic + API improvement
    api.ts        — OpenAI-compatible API calls
    learning.ts   — Pattern detection for rule suggestions
    responseAudit — Local + API response scoring
    reports.ts    — Aggregates events into UsageReport
    storage.ts    — chrome.storage.local wrapper
    types.ts      — All TypeScript types

---

Development

npm run dev       # watch mode (rebuilds on change)
npm run build     # production build
npm run type-check  # TypeScript checks without building

After npm run dev or npm run build, reload the extension in chrome://extensions to pick up changes.

---

Build Phases

• Phase 1 (this): Local MVP — heuristics, rules, dashboard, no backend
• Phase 2: API-enhanced rewriting + response audit improvements
• Phase 3: Team features, optional backend, shared rules
• Phase 4: Enterprise — SSO, RBAC, audit logs, SOC2 path