🏡 Gaff

Collaborative UK rental hunting for two people sharing one search.

Gaff scrapes the big letting portals, spots the same flat listed across all of them, and adds the facts that actually decide a tenancy: commute, council tax, broadband, energy rating. Then two people swipe through the results and build a shared shortlist. When you both keep the same place, it becomes a match.

Note

Archived personal project. Gaff was built for one real two-person house hunt. The hosted instance and its infrastructure have been retired, so this repo is just the source, kept public as a reference. It's parameterised so you can run your own (see Deploy).

What it does

House-hunting as a couple usually means two browsers, three portals, the same flat listed three times at three prices, and no shared memory of "did we both like that one?". Gaff turns all of that into a single queue.

• One feed across portals. Rightmove, Zoopla and OpenRent are scraped on a schedule. The same property, listed and priced differently on each, is grouped into one card.
• Decisions, not tabs. Review one listing at a time with keyboard shortcuts (keep, skip, defer). A skip is yours alone. A keep that you both make becomes a match.
• The facts up front. Every listing carries commute times to your stations, council tax band, broadband availability, energy rating, nearby amenities, and a short written read of what stands out and what to watch for.
• A shared shortlist. Matches flow into a pipeline: both kept, in conversation, viewing booked, offer placed. Two people track one hunt without a spreadsheet.
• A daily digest. When a search finishes scraping, each of you gets an email of what's new and waiting to review.

Screens

Review the queue, filter by beds, price, commute, EPC	Shortlist pipeline of mutual matches
Searches on independent schedules	Compare two listings, true monthly cost

Listing detail

Every clustered property opens to a full dossier: photo gallery, cross-portal price tracking, a map with transit times, and the public records (EPC, broadband, council tax, amenities) pulled in by the enrichment pipeline.

Photo gallery	Price, council tax & map

How it works

flowchart LR
    S[Searches] -->|scheduled| SC[Scrape<br/>Rightmove · Zoopla · OpenRent]
    SC --> CL[Cluster and de-dupe<br/>across portals]
    CL --> EN[Enrich]
    EN --> AIN[Claude: highlights and watch-outs]
    EN --> GE[Commute · transit · amenities]
    EN --> PR[EPC · broadband · council tax]
    AIN --> Q[Review queue]
    GE --> Q
    PR --> Q
    Q -->|keep / skip / defer| M{Both kept?}
    M -->|yes| PL[Shortlist pipeline]
    Q -->|new since last run| DG[Daily digest email]

Loading

Scraping, clustering and enrichment run as Trigger.dev jobs. The web app just reads the results. The review queue and the digest share one selection module, so the email can never promise listings the queue would then filter out.

Stack

• TanStack Start (React 19) - full-stack web framework, the app itself
• Cloudflare Workers - serverless runtime the app is deployed to
• Neon + Drizzle - serverless Postgres and the type-safe ORM
• Trigger.dev - background jobs: scrape, cluster, enrich, email
• Better Auth behind Cloudflare Access - sign-in and the Zero Trust gate
• Cloudflare R2 - object storage for cached, resized listing photos
• Claude - writes each listing's highlights and watch-outs
• Resend + React Email - the match and digest emails
• Doppler - secret storage and runtime injection
• Pulumi - Cloudflare infrastructure as code (KV, R2, DNS, Access)
• Bun · Biome · Vitest - package manager, linter, tests

Develop

bun install
bash scripts/setup-doppler.sh   # Doppler project `gaff`, config `dev`
bun run dev                     # web + Trigger.dev dev server
bun run db:migrate              # migrate your Neon branch

See .env.example for every variable the app reads.

🚀 Deploy your own

Gaff runs on free tiers of Cloudflare, Neon and Trigger.dev. Standing up a fresh instance means creating a few accounts, dropping their keys into Doppler, provisioning the Cloudflare resources with Pulumi, and pushing.

Full walkthrough: docs/DEPLOY.md. The short version:

# 1. Accounts: Cloudflare, Neon, Trigger.dev, Doppler, Anthropic, Google Maps,
#    Zyte, EPC Open Data, Resend (docs/DEPLOY.md says what each is for)
# 2. Put their keys in Doppler (gaff/dev + gaff/prd); checklist in .env.example
# 3. Set your own domain, org and IDs ("Make it yours" in DEPLOY.md)
t-stack provision      # Cloudflare KV, R2, DNS, Access via Pulumi
bun run db:migrate:prod
bun run deploy         # build + wrangler deploy

License

MIT.