PromptVerse

PromptVerse is an open-source workspace for discovering, organizing, and transforming AI prompts. It combines a polished Next.js interface with a SQLite-backed API so teams can build, iterate, and share prompt libraries with confidence.

✨ Highlights

• Rich prompt management – curate prompts with categories, tags, favorites, and version history.
• Database-backed workflows – create, edit, clone, and rate prompts stored in a shared SQLite catalog.
• Lightning-fast discovery – advanced filtering and live search keep the right prompt a keystroke away.
• Built-in transformer – open the transformer window from anywhere with Ctrl + Alt + Space.
• ChatGPT supercharge – flip the transformer to ChatGPT mode for guided mode suggestions, inline analytics, and developer-grade scaffolds.
• Offline-ready PWA – install PromptVerse as a desktop experience on Windows 11 or any modern browser.
• Dark mode design system – accessible theming tuned for productive day and night work.

🚀 Getting Started

# Install dependencies
npm install

# Local offline dev: create .env.local with a SQLite path
echo "DATABASE_URL=file:./db/dev.db" > .env.local

# Seed local SQLite and create tables (applies schema automatically)
npm run db:seed:local

# Start local development (Node + Next.js runtime)
npm run dev:local

Open http://localhost:3000 and explore the seeded workspace. You can change the SQLite path by updating DATABASE_URL in .env.local.

npm run dev:local uses scripts/dev-local.js, which temporarily flips every API route to runtime = "nodejs", launches next dev, and automatically restores runtime = "edge" as soon as the dev server exits (even on Ctrl+C). That keeps the working tree clean without manual toggles.

Need to inspect or automate the runtime swap yourself? Run node scripts/toggle-runtime.js --help for options such as --dry (preview) and --backup.

🤖 Optional: Local AI Proxy (LongCat / OpenAI-compatible)

If you have a LongCat (or any OpenAI-compatible) API key you can run the bundled proxy so the app always targets http://127.0.0.1:8787/v1 during local development. The proxy adds request caching, rate-limit jitter/backoff, model remapping, and optional shared-secret enforcement.

# 1) Configure credentials
echo "LONGCAT_KEY=lc-..." >> .env.local
# Optional shared secret so only your app can call the proxy
echo "PROXY_TOKEN=my-shared-secret" >> .env.local
# Override the default model so requests map to LongCat
echo "AI_MODEL=LongCat-Flash-Chat" >> .env.local

# 2) Start dev (proxy launches automatically if LONGCAT_KEY is set)
npm run dev:local

# Or launch the proxy manually in another terminal
npm run ai:proxy

When the proxy autostarts, scripts/dev-local.js injects AI_BASE_URL=http://127.0.0.1:<port>/v1 (default port 8787). If you define PROXY_TOKEN, the proxy enforces it and the Next.js app forwards it via AI_API_KEY, so every call carries the required Authorization header.

Key proxy environment variables (optional unless noted):

Variable	Purpose	Default
LONGCAT_KEY	LongCat API token used for upstream requests	- (required to start proxy)
LONGCAT_BASE	Upstream OpenAI-compatible base URL	https://api.longcat.chat/openai/v1
AI_MODEL	Default model forwarded to the upstream API	LongCat-Flash-Chat
PROXY_TOKEN	Shared secret required by the proxy	unset (no auth)
AI_PROXY_PORT	Port bound by the proxy	8787
AI_PROXY_HOST	Host bound by the proxy	127.0.0.1

In production you can deploy the proxy as a standalone Node service (or point AI_BASE_URL directly at your provider) and set AI_BASE_URL / AI_API_KEY accordingly.

🌱 Seed Baseline Data

Run one of the following to load the default user (Hany alsamman <[email protected]>), engineering categories, models, and starter templates:

npm run db:seed:local   # local SQLite seed (applies schema and data)
npm run db:seed         # uses scripts/seed.ts (better-sqlite3)
npx prisma db seed      # uses prisma/seed.ts (Prisma Client)

Both commands produce the same baseline dataset so the transformer’s template gallery and model picker are ready to go.

Set SEED_USER_EMAIL / SEED_USER_NAME in .env to customise the default account used during seeding.

☁️ Cloudflare Preview & Deploy

PromptVerse supports Cloudflare Pages with the edge runtime and D1. Preview and production use the same remote D1 and KV bindings configured in wrangler.toml.

# Preview against remote D1/KV (deploys a Pages preview on branch "preview")
npm run preview

# Deploy to production
npm run deploy

• Configuration lives in wrangler.toml (top-level and [env.preview] bindings).
• API routes run on the edge in production (literal runtime = "edge") and the dev wrapper temporarily switches them to Node/SQLite when you run npm run dev:local.
• KV binding name is promptverse_kv; D1 binding name is promptverse_d1_db.

🔁 Seeding Prompts on Cloudflare

Set an ADMIN_SEED_TOKEN secret in Cloudflare Pages (and optionally your GitHub environment). Then deploy and run:

curl -X POST https://your-domain/api/admin/seed-prompts \
  -H 'x-seed-token: your-secret'

The endpoint reuses the same logic as the local seed script but talks to D1.

GitHub Actions / CI secrets

Wrangler needs credentials in CI—local wrangler login state isn’t available in GitHub runners. Store these as repository secrets and expose them before running npm run preview or npm run deploy:

Secret	Purpose
CLOUDFLARE_ACCOUNT_ID	Your Cloudflare account id (15576ba2b25a9f04dcc0b8dcd5c0886a).
CLOUDFLARE_API_TOKEN	API token with Pages:Edit, D1:Edit, and Workers KV:Edit scopes.
ADMIN_SEED_TOKEN	Shared secret required by /api/admin/seed-prompts (optional).

Example workflow snippet:

env:
  CLOUDFLARE_ACCOUNT_ID: ${{ secrets.CLOUDFLARE_ACCOUNT_ID }}
  CLOUDFLARE_API_TOKEN: ${{ secrets.CLOUDFLARE_API_TOKEN }}

If these variables are missing you will see Cloudflare API authentication errors such as code: 10000 during deploy.

🔌 Configure the Transformer Provider

The /api/transform-prompt endpoint speaks to any OpenAI-compatible API, including hosted OpenAI models and local runtimes such as Ollama. Set the following environment variables in your .env file to choose a provider:

# Example: local Ollama
AI_BASE_URL="http://localhost:11434/v1"
AI_MODEL="llama3"

# Example: OpenAI (paid)
AI_BASE_URL="https://api.openai.com/v1"
AI_API_KEY="sk-..."
AI_MODEL="gpt-4o-mini"

# Example: Longcat (free tier)
AI_BASE_URL="https://api.longcat.chat/v1"
AI_API_KEY="lc-..."    # obtain at https://longcat.chat/platform/docs
AI_MODEL="LongCat-Flash-Chat"
# Other option: LongCat-Flash-Thinking

AI_API_KEY is optional for self-hosted providers that do not require authentication. PromptVerse automatically normalizes the base URL, so both local and remote deployments work without code changes.

Using Longcat's free tier

1. Register at LongCat's platform and complete the sign-up form.
2. After logging in, open the API Keys page.
3. Your account automatically includes a default key—copy its value into AI_API_KEY.
   • Name: application identifier.
   • Key: secret token—store securely.

💬 ChatGPT Mode

The transformer now includes a ChatGPT toggle designed for developer workflows. Selecting this mode swaps in a system prompt that enforces a structured scaffold:

• Top-level directives (/ROLE, /TASK, /FORMAT, /CONTEXT, /QUALITY BAR, /ASK).

• Optional reasoning add-ons (/STEP-BY-STEP, /CHAIN OF THOUGHT, /CHECKLIST, /PM MODE, and more) that are only appended when the model detects they will add value.

• Follow-up questions focused on clarifying ambiguous requirements.

• Guided mode selection helps you pin modes (e.g. /STEP-BY-STEP) based on live suggestions derived from the raw brief.

• Inline analytics tracks the most-used modes so you can see what the team relies on while iterating prompts.

• Copy prompts as Markdown, RTF, or open a ChatGPT tab preloaded with the scaffold—the tool also copies the prompt to your clipboard on the way.

• Saving from the transformer auto-generates a crisp title and description, pulling /ROLE + /TASK metadata in ChatGPT mode or summarising the opening brief in Standard mode.

• Engineering template gallery (inspired by OpenAI's "ChatGPT for engineering teams") offers ready-to-tailor prompts for reviews, architecture decisions, debugging, migrations, and incidents. Templates adapt automatically to Standard or ChatGPT mode.

• Save finished prompts directly to your workspace—Markdown responses are auto-titled and tagged before persisting.

• Launch an embedded tour with Ctrl + H to learn the layout (templates, raw prompt builder, guided modes, analytics, and actions).

The output is always Markdown so you can paste it directly into ChatGPT or any OpenAI-compatible chat interface. Switch back to Standard mode to regain JSON exports.

📦 Install as an App

1. Launch PromptVerse in Microsoft Edge, Chrome, or another PWA-capable browser.
2. Click Install App in the top toolbar (or use your browser's install option).
3. Follow the prompts to pin PromptVerse to your Start menu or taskbar.

The install experience uses a Progressive Web App manifest and service worker, so PromptVerse keeps working even when your network drops.

🧠 Core Features

• Prompt library – read, favorite, and rate prompts, with author attribution and usage stats.
• Advanced search – filter by category, model, tags, rating, favorites, and date range.
• Prompt creation – capture structured metadata and persist it instantly to the database.
• Version manager – clone prompts, record version notes, and evolve prompts safely.
• Prompt transformer – send prompts through the /api/transform-prompt endpoint for automated refinement.
  • Switch between the default profile and a ChatGPT-targeted profile that outputs /ROLE, /TASK, /ASK, and smart reasoning modes such as /CHAIN OF THOUGHT or /CHECKLIST when relevant.
  • Guided mode suggestions, pinned modes, and inline analytics keep ChatGPT prompts sharp and measurable.
  • Engineering template library with ready-to-tailor prompts for code reviews, architecture debates, debugging plans, migrations, and incident postmortems.
  • Save transformed prompts (Markdown or JSON) straight into your library with generated titles, sensible defaults, and tags.
  • Interactive tour (Ctrl + H) walks new teammates through the transformer workflow step by step.

🛠️ Tech Stack

• Framework: Next.js 15 (App Router) + TypeScript 5 + React 19
• Styling: Tailwind CSS 4 + shadcn/ui + Lucide icons
• State & Forms: React Hook Form, Zustand, TanStack Query
• Backend: SQLite (local dev) + D1 (Cloudflare) via a unified async data layer, REST APIs, Socket.IO-ready server bootstrap
• Content Tools: MDX Editor, React Markdown, Syntax Highlighter
• Testing & Quality: ESLint, TypeScript strictness, automated builds

🧭 Project Structure

src/
├── app/                 # App Router routes, API handlers, and layout
├── components/          # Reusable UI, including shadcn/ui primitives
├── hooks/               # Custom React hooks
├── lib/                 # Utilities, SQLite data helpers, socket helpers
└── scripts/             # Database seed utilities

🤝 Contributing

1. Fork the repository and create a branch: git checkout -b feature/amazing-idea
2. Install dependencies and ensure the database is running.
3. Run npm run lint and npm run build before submitting your pull request.
4. Open a PR describing your changes and link any relevant issues.

We welcome issues, feature ideas, and documentation updates from the community.

📄 License

PromptVerse is released under the MIT License. Feel free to use it in your own projects or adapt it for your team.

🌐 Community

Have a question or want to share how you're using PromptVerse? Open a discussion or join the conversation in the issue tracker. Let's build a richer prompt ecosystem together!