Build a Tweet SaaS with Next.js, Prisma Postgres, and Ollama

ollama pull gemma3:4b

ollama run gemma3:4b

curl http://localhost:11434/api/tags

{

  "models": [{

    "name": "gemma3:4b",

    "family": "gemma3",

    "parameter_size": "4.3B",

    "quantization_level": "Q4_K_M"

  }]

}

cd tweetsmith

I have a fresh Next.js 15 project with Tailwind CSS already set up.

I need you to create a minimalist single-page UI for a TweetSmith app called "TweetSmith". 

**Design requirements:**

- Dark theme only (no light/dark toggle)

- Typefully-inspired aesthetic: sophisticated, clean, minimal

- Color palette:

  - Background: #141414 (soft charcoal)

  - Card/inputs: #1c1c1c

  - Borders: #2a2a2a

  - Muted text: #888888

  - Foreground/text: #fafafa

- Typography: Geist font (already configured), small refined sizes

- Labels should be uppercase with letter-spacing

**What to create:**

1. Update `app/globals.css` with the dark color palette and CSS variables

2. Update `app/layout.tsx` with proper metadata (title: "TweetSmith") and force the dark background

3. Create `app/components/TweetTransformer.tsx` - a client component with:

   - A textarea input for draft tweets

   - A character counter (X / 280)

   - A "Transform" button (disabled when empty)

   - An output section for the result (only visible when there's a result)

   - Loading state ready for future API integration

   - The handleTransform function should just console.log for now

4. Update `app/page.tsx` with:

   - Simple header with app name and tagline

   - The TweetTransformer component

   - Subtle footer saying "powered by ollama"

Keep it minimal, no extra features, just clean functional UI. Maximum width should be max-w-md for a focused feel.

git add .

git commit -m "Add TweetSmith UI"

Create `app/lib/ollama.ts` - a helper file to communicate with a local Ollama LLM.

Requirements:

- Ollama runs at http://localhost:11434

- Model name: "gemma3:4b"

- Use the /api/generate endpoint

- Create TypeScript types for OllamaRequest and OllamaResponse

- Export a function `generateWithOllama(prompt: string)` that:

  - Sends a POST request with the prompt

  - Uses stream: false (no streaming, keep it simple)

  - Returns the response text as a string

  - Throws an error if the request fails

Keep it minimal with clear comments explaining what each part does.

Create `app/api/transform/route.ts` - a Next.js API route that transforms tweets.

Requirements:

- POST endpoint that accepts { draft: string } in the body

- Validate that draft exists and is a string (return 400 if invalid)

- Use the generateWithOllama function from "@/app/lib/ollama"

- Build a prompt that tells the LLM to:

  - Act as a tweet formatter

  - Make the draft cleaner, more engaging, well-formatted

  - Keep it under 280 characters

  - Return only the improved tweet, nothing else

- Return { transformed: string } on success

- Return { error: string } with status 500 if Ollama fails

- Add a helpful error message asking if Ollama is running

Use NextRequest and NextResponse from next/server.

Update the TweetTransformer component to call the transform API:

1. Add an "error" state (useState) to handle errors

2. Update handleTransform to:

   - Reset error and result states first

   - Set loading to true

   - Call POST /api/transform with { draft: draftTweet }

   - On success: set the transformed tweet from response

   - On error: set error message from the catch

   - Use finally to always set loading to false

3. Display error message in red below the button when there's an error

Keep the existing UI structure, just wire up the real API call.

Update the ollama.ts file to add temperature for response variety:

1. Add an optional "options" field to OllamaRequest type with temperature?: number

2. In the request body, add:

   options: {

     temperature: 0.7

   }

git add .

git commit -m "Connect to Ollama LLM for tweet transformation"

Create `app/components/FilterOptions.tsx` with filters for tweet generation.

Requirements:

- Export a `Filters` type with: maxChars (number) and emojiMode ("none" | "few" | "many")

- Export two components:

  1. `FilterButton` - a toggle button showing current filter values (e.g., "Filters 280 · few")

  2. `FilterPanel` - the expanded controls panel

FilterButton props: isOpen, onToggle, filters

FilterPanel props: filters, onFiltersChange

FilterPanel should include:

- A range slider for maxChars (100-280, step 20)

- Three icon buttons for emoji mode using Lucide icons:

  - Ban icon for "none"

  - Smile icon for "few"  

  - SmilePlus icon for "many"

- Display inline with dividers between sections

- Compact design with small buttons (h-7 w-7)

Use SlidersHorizontal and ChevronDown from lucide-react for the button.

Style: dark theme, rounded-lg, border-border, bg-card.

Create `app/components/ContextSettings.tsx` with context management.

Export two components:

1. `ContextButton` - toggle button with User icon from lucide-react

   - Props: isOpen, onToggle, hasContext (boolean)

   - Show a small dot indicator when context is set

2. `ContextPanel` - just the textarea

   - Props: onContextChange

   - Load/save to localStorage with key "TweetSmith-context"

Use User and ChevronDown icons from lucide-react.

Keep the textarea at 2 rows, placeholder about style/tone.

Update `app/components/TweetTransformer.tsx` to use the new component structure.

Changes:

1. Import ContextButton, ContextPanel from ContextSettings

2. Import FilterButton, FilterPanel, Filters from FilterOptions

3. Add state for which panel is open: type OpenPanel = "none" | "context" | "filters"

4. Track hasContext state (check localStorage on mount)

Layout structure:

- Settings row: flex container with gap-2 containing both buttons INLINE

- Below the row: conditionally render either ContextPanel or FilterPanel (only one at a time)

- Clicking one panel closes the other

This keeps both buttons always on the same line, with expanded content appearing below.

Default filters: maxChars 280, emojiMode "few"

Update `app/api/transform/route.ts` to use the filter values.

Key changes:

1. Extract filter values directly: maxChars and emojiMode with defaults (280, "few")

2. Build emoji rule as a simple string based on mode

3. Put STRICT LIMITS at the TOP of the prompt (most important):

   - "Maximum {maxChars} characters (THIS IS MANDATORY)"

   - Emoji rule

   - "No hashtags"

4. Add context as "Author style:" if provided

5. Keep GUIDELINES brief: lead with value, sound human, be engaging

6. End with: 'Respond with ONLY the rewritten tweet. No quotes, no explanation.'

Remove any complex buildFilterRules function - inline everything for clarity.

The prompt should be shorter and more direct for better local LLM compliance.

git add .

git commit -m "Add filter options and context settings"

Create `app/components/TweetPreview.tsx` - a tweet-like preview card with 3 states.

Props: content (string | null), isLoading (boolean)

Structure:

1. Header with profile image, name, verified badge, and handle

2. Content area (changes based on state)

3. Footer with character count or placeholder

Three states:

1. EMPTY (no content, not loading):

   - Show placeholder text styled like a tweet but in text-muted

   - Example: "Follow us on X to stay updated on all the latest features and releases from Prisma! 🚀\n\nYour polished tweet will appear here ✨"

   - Footer shows "prisma.io"

2. LOADING (isLoading true):

   - Show 3 animated skeleton bars with animate-pulse

   - Different widths: 90%, 75%, 60%

   - Footer skeleton bar

3. CONTENT (has content):

   - Show the actual tweet text

   - Copy button in header (using Copy/Check icons from lucide-react)

   - Footer shows "X / 280 characters"

Use Image from next/image for the profile picture.

Add verified badge as inline SVG (Twitter blue checkmark).

Update `app/page.tsx` to use a logo image instead of text for the header.

Changes:

1. Import Image from "next/image"

2. Replace the h1 text with an Image component:

   - src="/icon-logo.png" (or your logo file)

   - width={80} height={80}

   - Add className="mb-3" for spacing

3. Keep the tagline text below: "polish your tweets with AI"

4. Use flex flex-col items-center on the header

The header should now show: Logo image centered, tagline below.

git add .

git commit -m "Add tweet preview card and logo"

generator client {

  provider = "prisma-client"

  output   = "../app/generated/prisma"

}

datasource db {

  provider = "postgresql"

}

model SavedTweet {

  id          String   @id @default(cuid())

  original    String   // The draft tweet input

  transformed String   // The polished/transformed tweet

  context     String?  // Optional user context/style used

  imageUrl    String?  // Optional image URL

  imageAlt    String?  // Optional alt text for accessibility

  createdAt   DateTime @default(now())

  updatedAt   DateTime @updatedAt

}

import { PrismaClient } from "../app/generated/prisma/client"

import { PrismaPg } from "@prisma/adapter-pg"

const adapter = new PrismaPg({

  connectionString: process.env.DATABASE_URL!,

})

const globalForPrisma = global as unknown as { prisma: PrismaClient }

const prisma = globalForPrisma.prisma || new PrismaClient({

  adapter,

})

if (process.env.NODE_ENV !== "production") globalForPrisma.prisma = prisma

export default prisma

{

  "scripts": {

    "dev": "next dev",

    "build": "next build",

    "start": "next start",

    "lint": "eslint",

    "db:test": "tsx scripts/test-database.ts",

    "db:studio": "prisma studio"

  }

}

import "dotenv/config"

import prisma from "../lib/prisma"

async function testDatabase() {

  console.log("🔍 Testing Prisma Postgres connection...\n")

  try {

    console.log("✅ Connected to database!")

    console.log("\n📝 Creating a test saved tweet...")

    const newTweet = await prisma.savedTweet.create({

      data: {

        original: "just shipped a new feature, its pretty cool i think",

        transformed: "Just shipped a new feature! 🚀 Pretty excited about this one ✨",

        context: "Tech founder, casual tone",

      },

    })

    console.log("✅ Created saved tweet:", newTweet)

    console.log("\n📋 Fetching all saved tweets...")

    const allTweets = await prisma.savedTweet.findMany()

    console.log(`✅ Found ${allTweets.length} saved tweet(s)`)

    console.log("\n🎉 All tests passed! Your database is working perfectly.\n")

  } catch (error) {

    console.error("❌ Error:", error)

    process.exit(1)

  }

}

testDatabase()

Create `app/api/tweets/route.ts` with GET, POST, and DELETE handlers.

Requirements:

- GET: Fetch all saved tweets ordered by createdAt desc

- POST: Save a new tweet with { original, transformed, context?, imageUrl?, imageAlt? }

- DELETE: Delete a tweet by id (passed as query param ?id=xxx)

Use try-catch blocks and return appropriate error responses.

Import prisma from "../../../lib/prisma"

# Save a tweet

curl -X POST http://localhost:3000/api/tweets \

  -H "Content-Type: application/json" \

  -d '{"original":"test draft","transformed":"Test polished! ✨"}'

# Get all tweets

curl http://localhost:3000/api/tweets

git add .

git commit -m "Add Prisma Postgres database"

Add a Save button to save transformed tweets to the database.

Requirements:

- Add a Save button next to the Copy button in TweetPreview

- Use POST /api/tweets with { original, transformed, context }

- Pass original (draft) and context from TweetTransformer to TweetPreview

UX States:

- Default: Bookmark icon with hover scale effect

- Saving: Spinning Loader2 icon + "Saving" text

- Saved: Green tinted background (emerald-500/10), checkmark icon with zoom animation

Important: The "Saved" state must persist until a NEW tweet is generated. Use useRef to track previous content and useEffect to reset saved state only when content changes. Do not use setTimeout to reset the saved state.

Match the minimal dark aesthetic of the app (200ms ease-out transitions, subtle hover states).

Add a Library feature to browse and reuse saved tweets.

Components to create:

- LibraryButton.tsx - Fixed top-right button with count badge

- LibraryPanel.tsx - Slide-in panel from right (360px, backdrop blur)

- SavedTweetCard.tsx - Tweet cards that look like published tweets

LibraryButton:

- Fixed position top-right (fixed top-6 right-6)

- Shows saved tweets count as badge

- Toggles panel open/close

LibraryPanel:

- Slides in from right with 300ms ease-out animation

- Backdrop overlay with blur

- Header with title, count, and close button

- Scrollable list of SavedTweetCard components

- Empty state with icon when no tweets saved

- Fetches tweets from GET /api/tweets when opened

SavedTweetCard:

- Looks like a real published tweet (profile image, name, verified badge, handle, date)

- Shows transformed tweet content only (not original)

- Footer: character count on left, Copy/Delete icons on right (subtle, brighten on hover)

- Click anywhere on card → loads transformed text into Draft textarea and closes panel

- Delete shows inline confirmation (Cancel/Delete buttons), not a modal

Integration:

- Add library state to TweetTransformer (isOpen, count)

- Fetch count on mount and after saving

- Pass onUseAsDraft callback to set draft and clear transformed tweet

- Refresh count when panel closes (in case tweets were deleted)

Styling: Match minimal dark aesthetic - subtle borders, muted colors, smooth 200ms transitions.

git add .

git commit -m "Add tweet library with save/browse/delete"

Add a theme system to my app with 3 dark themes and a minimal theme switcher.

THEMES:

1. "Disco" - Purple/violet twilight vibes

   - Background: #17171c (deep blue-black)

   - Accent: #a78bfa (soft violet)

   

2. "Dust" - Desert warmth, amber tones

   - Background: #1a1816 (warm charcoal)

   - Accent: #d4a574 (warm amber/sand)

   

3. "Press" - Old newspaper, pure greys

   - Background: #262626 (true grey)

   - Accent: #a3a3a3 (neutral grey)

IMPLEMENTATION:

- Use CSS custom properties (:root and [data-theme="..."]) for all colors

- Add a subtle radial gradient glow at the top of the page using the accent color

- Create a ThemeSwitcher component with small colored dots (one per theme)

- Place the switcher in the footer for minimal UI impact

- Persist theme choice in localStorage

- Add smooth transitions when switching themes (0.3s ease)

- Prevent transition flash on page load with a "no-transitions" class

UX REQUIREMENTS:

- Each dot shows the accent color of that theme

- Selected theme has a subtle ring + slight scale up

- Unselected themes are dimmed (opacity 40%) and brighten on hover

- Theme changes should animate smoothly across all UI elements

Add smooth animated collapsible panels that expand/collapse without layout jumps.

ANIMATED PANEL COMPONENT:

Create a reusable AnimatedPanel component using the CSS Grid trick for height animation:

- Use display: grid with gridTemplateRows

- Closed state: gridTemplateRows: "0fr" (collapses to 0 height)

- Open state: gridTemplateRows: "1fr" (expands to content height)

- Wrap children in a div with overflow: hidden

- Add opacity fade: 0 when closed, 1 when open

- Transition: "grid-template-rows 0.25s cubic-bezier(0.32, 0.72, 0, 1), opacity 0.2s ease"

COLLAPSE GAP TRICK:

If parent uses gap/space between items, add negative margin when closed to collapse the gap:

- marginTop: isOpen ? undefined : "-12px" (adjust based on your gap size)

- Animate the margin too for smooth effect

PREVENT SCROLLBAR LAYOUT SHIFT:

Add to your global CSS on html element:

- scrollbar-gutter: stable (reserves space for scrollbar)

- overflow-x: hidden (prevents horizontal scroll)

This creates buttery smooth expand/collapse without the jarring height jump or scrollbar layout shift.

git add .

git commit -m "Add themes and smooth animations"

UPLOADTHING_TOKEN=your_token_here

Add image upload to tweets using UploadThing.

Requirements:

1. Users should be able to attach ONE image to their tweet

2. The image should only be uploaded to UploadThing when clicking "Save" (not when selecting the image)

3. While editing, show a local preview using URL.createObjectURL() - this avoids orphaned uploads if the user changes their mind

4. Show upload progress in the Save button ("Uploading..." → "Saving...")

5. Allow removing the selected image before saving (X button on the image preview)

6. Display saved images in the tweet library/cards

Implementation steps:

1. Create UploadThing FileRouter at app/api/uploadthing/core.ts with a "tweetImage" route (4MB max, 1 file)

2. Create the route handler at app/api/uploadthing/route.ts

3. Create typed utilities at app/lib/uploadthing.ts with useUploadThing hook

4. Update TweetPreview component:

   - Add file state (File object) and preview URL state

   - Add hidden file input + "Add image" label/button

   - Show image preview with remove button

   - In handleSave: if file exists, call startUpload() first, then save tweet with the returned URL

5. Update SavedTweetCard to display imageUrl if present

The schema already has imageUrl and imageAlt fields.

Key pattern: Store File locally → preview with createObjectURL → upload only on save → save URL to database

git add .

git commit -m "Add image upload with UploadThing"