# Prompt Tuner

A self-hosted, privacy-first batch testing tool for LLM prompts.

Write a parameterized prompt once. Supply hundreds of variable combinations via CSV. Run them all against Gemini or Groq — in parallel — and compare outputs side-by-side in a spreadsheet-style view.

---

✨ Features

• Parameterized Templates — Write prompts using {variable} syntax (e.g., Summarize {topic} in a {tone} tone).
• Manual Mode — Fill in one set of values, add to the test table, run on demand.
• CSV Batch Mode — Upload a CSV; each row becomes a test case. One click runs all of them.
• Concurrent Execution — Up to 3 LLM calls run in parallel, maximizing throughput without triggering rate limits.
• Multi-Key Round-Robin — Paste multiple API keys (one per line). The backend distributes requests across them using a thread-safe rotating index.
• Inline Editing — Edit any test row's variable values directly in the table without re-uploading.
• Export to CSV — Download all results (inputs + outputs + metadata) as a CSV for offline analysis.
• Privacy-First — API keys and prompt content are never written to server logs, files, or a database. Everything is processed in-memory.

---

🚀 Quick Start

Local Development

# 1. Clone the repository
git clone https://github.com/SamratRay2005/Prompt_tuner
cd Prompt_tuner

# 2. Install dependencies
pip install flask

# 3. Run
python3 app.py

Open http://localhost:5005 in your browser.

Deploy to Vercel (One Click)

Click the button above, or follow these steps manually:

# Install the Vercel CLI
npm i -g vercel

# From the project root
vercel

Follow the prompts. Vercel auto-detects the vercel.json configuration and deploys the Flask app as a serverless function. No extra configuration needed.

---

📖 Usage Guide

See the in-app guide for full step-by-step instructions with visuals.

Quick reference:

1. Select a Provider (Gemini or Groq) and paste your API Key(s).
2. Choose a Model from the dropdown (auto-fetched from the provider).
3. Write your Prompt Template using {variable} placeholders.
4. Choose Manual (single run) or CSV (bulk run) mode.
5. Click "Add to Test Table" — rows appear with idle status.
6. Click "Run All" or select specific rows and click "Run Selected".
7. Click any output cell to read the full response. Click "Export CSV" to download all results.

---

🏗️ Architecture

prompt_tuner/
├── app.py                      # Flask thin controller (routing only)
├── providers/
│   ├── base.py                 # LLMProvider ABC + Value Objects
│   ├── groq_provider.py        # Groq Strategy implementation
│   ├── gemini_provider.py      # Gemini Strategy implementation
│   ├── registry.py             # ProviderFactory (Open/Closed)
│   ├── key_rotator.py          # Thread-safe Singleton key rotation
│   └── http_client.py          # urllib wrapper (no third-party HTTP lib)
├── templates/
│   ├── index.html              # Main app page
│   └── help.html               # Usage guide page
├── static/
│   ├── style.css               # Full CSS design system
│   └── script.js               # All frontend logic (Vanilla JS)
├── api/index.py                # Vercel serverless wrapper
└── vercel.json                 # Vercel routing + security headers

Design Patterns: Strategy, Factory, Template Method, Singleton, Value Object
SOLID: Each class has one reason to change. Adding a new provider requires no changes to existing files.

---

➕ Adding a New LLM Provider

1. Create providers/your_provider.py:

from providers.base import CompletionRequest, LLMProvider
from providers.http_client import HttpResponse, make_http_request

class YourProvider(LLMProvider):
    provider_name = "yourprovider"

    def _call(self, key: str, req: CompletionRequest) -> HttpResponse:
        # Build and fire the HTTP request
        ...

    def _parse_response(self, resp: HttpResponse) -> str:
        # Extract the generated text from the response body
        return resp.json()["your"]["nested"]["text"]

    def list_models(self, key: str) -> list[str]:
        # Fetch available model IDs
        ...

2. Register it in providers/registry.py:

_REGISTRY = {
    "groq":         GroqProvider,
    "gemini":       GeminiProvider,
    "yourprovider": YourProvider,   # ← one line
}

That's it. No other files change.

---

🔒 Security

Layer	Protection
API keys in POST body	Never appear in URLs, access logs, or browser history
Zero server-side logging of keys or prompts	Vercel Function Logs show only provider=X model=Y key_index=N
HTTP Security Headers	X-Frame-Options, CSP, X-Content-Type-Options, Referrer-Policy
Cache-Control: no-store	LLM responses are never cached by browsers or CDNs
No database	Zero persistence — everything lives in browser memory
credentials: same-origin	Fetch calls are strictly same-origin

---

🛠️ Tech Stack

Layer	Technology
Backend	Python 3 + Flask
HTTP Client	urllib (stdlib only, zero extra dependencies)
Frontend	Vanilla JavaScript (no framework)
CSV Parsing	PapaParse 5.4.1 (CDN)
Icons	Font Awesome 6.4.0
Fonts	Google Fonts (Inter, Outfit, JetBrains Mono)
Production	Vercel (serverless Python)

---

📄 License

MIT — free to use, modify, and distribute.