PCease

Plan a PC for the Indian market: compare part prices across Indian retailers, check compatibility as you build, and get advice from an AI that only recommends parts we actually list.

Live site: https://pcease-virid.vercel.app · Repository: https://github.com/adithya-jithesh/PCease

Pages

Page	Link	What it does
Home	/	Live sample build, categories, how it works
Parts	/parts	Search and filter the catalogue; each part page compares retailer prices and shows price history
Builder	/builder	Slot-by-slot builder with live compatibility and power checks
Compare	/compare	Up to four parts side by side
AI Advisor	/advisor	Chat with the AI advisor, or use the budget planner
Forum	/forum	Threads, replies, voting and attached builds
Sign in	/login	Email sign-in (Google appears once enabled in Supabase)
Dashboard	/dashboard	Your saved builds and threads (signed in)
Price admin	/admin/prices	Review price runs and edit prices (admins only)

Features

• Parts catalogue: search and filter CPUs, GPUs, motherboards, memory, storage, PSUs, cases and coolers. Every part page compares prices across retailers, highlights the cheapest, and charts the best price over time.
• Builder: pick parts slot by slot. Live checks cover sockets, memory generation, RAM slots, board/case fit, GPU length, cooler mounting and height, PSU headroom, display output, board tier and CPU/GPU balance. The part picker can hide anything incompatible with what you've already chosen.
• Share links: any build can be shared as a link, no account needed.
• Compare: put up to four parts side by side, with the better spec highlighted on each row.
• AI advisor (Gemini):
  • answers PC hardware questions only; a classifier turns away anything else (including attempts to override its instructions) before it reaches the main model
  • recommends parts through tools that search the catalogue, check compatibility and plan builds, so every suggestion is a real listing at today's price
  • mentioned parts become cards you can add to your build, and suggested builds load into the builder in one click
  • falls back across several Gemini models when one is busy, and retries replies that get cut off
• Budget planner: tries every CPU and GPU pairing, completes each with compatible parts, and keeps the strongest build within budget.
• Live prices: a daily job uses Gemini with Google Search to refresh prices at our retailers, with safety checks and full price history. Admins can review runs and correct prices by hand.
• Accounts (Supabase Auth): save builds, make them public or private, duplicate them, and see how each build's price has moved since you saved it.
• Forum: threads by topic, replies, voting, and attaching a saved build to a post.

Tech stack

Layer	Choice
App	Next.js 16 (App Router, Server Components, Server Actions), React 19, TypeScript
Styling	Tailwind CSS v4 with a navy and teal design-token theme
Data & auth	Supabase (Postgres, row-level security, Supabase Auth)
AI	Google Gemini API via @google/genai: function calling for the advisor, Google Search grounding for prices
Testing	Vitest
Hosting	Vercel (including Vercel Cron) + Supabase
CI	GitHub Actions: lint, typecheck, tests and build on every push

Getting started

1. Create a Supabase project in the Supabase dashboard.

2. Apply the schema and seed data. Open the project's SQL Editor and run these files in order:

   1. supabase/migrations/0001_init.sql
   2. supabase/migrations/0002_price_tracking.sql
   3. supabase/seed.sql (safe to re-run; it never overwrites existing prices)

3. Configure auth (dashboard → Authentication):

   • URL Configuration: set the Site URL and add http://localhost:3000/auth/callback plus https://<your-domain>/auth/callback to Redirect URLs (docs).
   • Email: Supabase's built-in email sender is heavily rate limited. Either turn Confirm email off (Sign In / Providers → Email) or connect a real provider under Emails → SMTP (docs); Resend has a free tier.
   • Google sign-in (optional): enable the Google provider with OAuth credentials from the Google Cloud console (guide). The button appears on the sign-in page automatically.

4. Set environment variables:

   cp .env.example .env.local

   Variable	Required	Where to get it
   NEXT_PUBLIC_SUPABASE_URL	yes	Supabase → Project Settings → API (docs)
   NEXT_PUBLIC_SUPABASE_PUBLISHABLE_KEY	yes	Supabase → Project Settings → API Keys → publishable key
   NEXT_PUBLIC_SITE_URL	recommended	Your site's URL, e.g. http://localhost:3000 locally
   GEMINI_API_KEY	for AI	Google AI Studio → API keys
   SUPABASE_SECRET_KEY	for prices	Supabase → Project Settings → API Keys → secret key. Server only; never share it.
   CRON_SECRET	for prices	Any long random string, e.g. node -e "console.log(require('crypto').randomBytes(24).toString('hex'))"
   ADMIN_EMAILS	for prices	Comma-separated emails allowed into /admin/prices

5. Run it:

   npm install
   npm run dev

   Then open http://localhost:3000.

AI models

The advisor, its topic check and the price lookups each try a list of Gemini models in order, moving to the next when one is overloaded (503) or out of quota (429), and skipping a busy model for a minute. Defaults live in src/lib/ai.ts. Google retires and renames models regularly (current models), so you can override the lists without a code change:

Variable	Used for	Needs
GEMINI_CHAT_MODELS	AI advisor	function calling
GEMINI_FAST_MODELS	off-topic check	JSON output
GEMINI_PRICE_MODELS	price updates	Google Search grounding

Each takes a comma-separated list, best first, e.g. GEMINI_CHAT_MODELS=gemini-3.8-flash,gemini-3.5-flash-lite.

Keeping prices current

/api/cron/prices refreshes the parts whose prices were checked longest ago (up to 30 per run). vercel.json schedules it daily, and Vercel authenticates it with CRON_SECRET. For each part, Gemini searches Google for the price at each of our retailers, and a price is only written when:

• it comes from one of our retailers, and Google Search returned evidence from that retailer's domain;
• any product link is on that retailer's domain;
• it's within 60–150% of the part's current median price (bigger jumps are left for a human to confirm).

Every change, whether automatic, manual or from the seed, is recorded in price_history, which drives the charts on part pages. Rejected prices and their reasons are logged per run and shown on /admin/prices, where admins can also refresh a part on demand or fix a price by hand.

Billing required for automatic prices. Free Gemini API keys get little or no Google Search grounding quota, so lookups fail with "quota exceeded" until billing is enabled on the key's Google Cloud project (billing, pricing, rate limits). The job detects this, stops early with a clear message and changes nothing. The AI advisor itself works on the free tier.

To refresh from your machine (reads .env.local):

npm run prices:update                  # the 10 stalest parts
npm run prices:update -- --limit 61    # everything
npm run prices:update -- rtx-4060      # specific parts

Deploying to Vercel

1. Import the repository in Vercel (framework preset: Next.js, root directory: the repo root, Node.js 22 or 24).
2. Add every variable from the table above under Settings → Environment Variables (docs), with NEXT_PUBLIC_SITE_URL set to your production URL.
3. Add https://<your-domain>/auth/callback to Supabase's Redirect URLs.
4. Push to main. Vercel deploys automatically, and the daily price job is registered from vercel.json.

From the command line with the Vercel CLI:

npx vercel link
npx vercel env add GEMINI_API_KEY production
npx vercel deploy --prod

Scripts

Command	What it does
npm run dev	Start the dev server
npm run build	Production build
npm run lint	ESLint
npm run typecheck	Generate route types and run tsc
npm test	Unit tests (compatibility, planner, advisor tools, model fallback, price validation, share links)
npm run prices:update	Refresh prices from the command line

Project layout

src/
  app/              routes: parts, builder, compare, advisor, forum, dashboard, auth, admin
    api/advisor/    streaming, tool-calling AI advisor
    api/cron/       scheduled price refresh
  components/       shared UI (header, part cards, build sheet, checks, charts, toasts)
  lib/
    ai.ts           Gemini client, model lists and fallback
    advisor/        advisor prompt, scope guard, tools and slug handling
    prices/         price lookup, validation and updater
    compat.ts       compatibility and power analysis
    planner.ts      budget build planner
    data.ts         server-side catalogue queries
    stores.ts       localStorage-backed build and compare state
    supabase/       browser, server and admin clients
  proxy.ts          session refresh and protected routes
scripts/            command-line tools
supabase/
  migrations/       schema, triggers and RLS policies
  seed.sql          retailers and starter catalogue

Troubleshooting

Problem	Fix
Sign-up says "email rate limit exceeded" or no confirmation email arrives	Supabase's built-in sender is limited to a few emails an hour. Turn off Confirm email or set up SMTP (see Getting started).
"Continue with Google" gives a 400 "provider is not enabled"	Enable the Google provider in Supabase. The button is hidden automatically until you do.
AI advisor says it isn't switched on	Set GEMINI_API_KEY and restart the server (or redeploy on Vercel).
AI errors mention a model "no longer available" (404)	Google retired the model. Set GEMINI_CHAT_MODELS / GEMINI_FAST_MODELS / GEMINI_PRICE_MODELS to current models.
"The AI is very busy right now"	Every configured model is overloaded (503). It usually clears within minutes; adding another model to the list helps.
Price updates stop with "Google Search quota is exhausted"	Enable billing on the Gemini key's project (see Keeping prices current).
CI fails at npm ci with "Missing: @emnapi/... from lock file"	Run npm install on any OS and commit package-lock.json. The @emnapi packages are pinned in devDependencies to keep Windows installs from dropping them.

Notes

• Seed prices are indicative. Run the price job, or edit prices on the admin page, before relying on them.
• Automatic prices are good but not perfect. Prices outside the safe range are skipped, so check the admin page now and then.
• The AI chat's rate limit and model cooldowns are held in memory per server instance. Use a shared store such as Upstash Redis if you need exact limits across instances.