PromptEQ
Tune your AI. Like an equalizer — but for how it talks.
PromptEQ is a visual custom-instructions builder. Pick an archetype, drag the faders, set your rules, and watch a sample reply and your instructions update live. When you're happy, polish them with AI and paste them into Claude, ChatGPT, Gemini, or Grok.
Dark homepage hero.
Generator with a prompt output in the terminal.
What It Does
Most people use AI on default settings — which is like using a speaker with no EQ. PromptEQ lets you dial in exactly how your assistant thinks, talks, and pushes back.
Four steps:
- Pick a Foundation — choose an archetype (Operator, Strategist, Learner, Creator); the faders sweep into place
- Dial the Mix — six vertical faders with five notches each
- About You — use cases, expertise, and optional role / current work / tools (never included in share links)
- Set Behavioral Rules — hard rules for uncertainty, ambiguity, filler, emoji, disagreement, plus your own
The Control Booth stays on screen the whole time. It shows a Default AI vs Your AI reply to a sample question, and the live instructions draft, which highlights each line as it changes. The draft is ready to paste as-is. Polish with AI rewrites it for your platform, then checks the result: every rule you set has to survive, emoji rules are enforced, and the text has to fit the platform's length limit. If the rewrite fails, it retries once with feedback; if it fails again, you keep the draft.
Every level of the Candor fader keeps an accuracy over agreement floor: the faders change how criticism is delivered, never whether the AI tells you you're wrong.
Archetypes
| Preset | Vibe | Good For |
|---|
| ⚡ Operator | Blunt. Fast. No fluff. | Decision-making, quick answers |
| ♟️ Strategist | Pushback and nuance. | Thinking partner, tradeoff analysis |
| 🔬 Learner | Patient. Step-by-step. | Learning new topics, research |
| 🎨 Creator | Riff mode. Loose structure. | Brainstorming, ideation |
Faders
| Fader | Range |
|---|
| Directness | Warm ↔ Blunt |
| Length | Minimal ↔ Deep |
| Candor | Gentle ↔ Stress-test |
| Analogies | Literal ↔ Analogy-first |
| Structure | Narrative Prose ↔ Bullets & Tables |
| Answer Order | Context First ↔ TL;DR First |
Behavioral Rules
Hard constraints that override the archetype defaults:
- When Uncertain — admit it openly, or give best guess
- Ambiguous Requests — ask for clarification, or assume and go
- Emoji Usage — never, sparingly, or freely
- Filler & Pleasantries — cut openers, sign-offs, and needless apologies
- Disagreement Style — flag concerns gently, or argue the other side hard
- Special Instructions — freeform custom rules for your workflow
Platform Support
Instructions are shaped per platform. ChatGPT gets two separate boxes, each with a live counter against its 1,500-character limit; the other platforms get one block with a suggested length.
- Claude — Settings > General > Profile > Custom Instructions
- ChatGPT — Settings > Personalization > Custom Instructions
- Gemini — Settings > Personal Intelligence > Instructions
- Grok — Settings > Customize > How would you like Grok to respond?
Tech Stack
- Next.js 16 (App Router) — full-stack React framework
- React 19 — component library
- Tailwind CSS — utility styling
- Phosphor Icons — icon system
- Grok API (xAI) — LLM polish step (
grok-4.20-0309-non-reasoning, ~7s, strict JSON-schema output)
- Railway — hosting (
npm run build, then npm start)
There is a backend: an API route that validates your settings, builds the prompt on the server (so the endpoint can't be used as an open LLM proxy), and calls xAI. It's rate-limited per IP (8 polishes a minute) and each model call times out after 20 seconds. No user data is stored.
Getting Started
git clone https://github.com/machovato/PromptEQ.git
cd PromptEQ
npm install
Create a .env.local file in the root:
XAI_API_KEY=your_xai_api_key_here
# Optional: override the polish model; set XAI_REASONING_EFFORT only for models that accept it
# XAI_MODEL=grok-4.3
# XAI_REASONING_EFFORT=low
Then run:
npm run dev
Open http://localhost:3000.
Run the tests (Node's built-in runner, no extra dependencies):
npm test
Contributing
PRs welcome. If you're pulling on this — hi 👋 — open an issue or just fork and go.
License
MIT
