nextup

Sign in as the owner, paste or upload screenshots of your streaming saved lists in append-only or full-update mode, confirm what was read from them, and see one deduplicated combined list — one row per title, a badge per service — that you can filter and sort and that never loses anything without asking you first.

What it does

Supported services: Netflix, Max, Prime Video, Disney+, Apple TV+, Paramount+, Starz and Peacock. Imports are owner-uploaded screenshots, not connections to streaming accounts.

Your saved watchlists are trapped inside separate streaming apps, so answering "what have I saved that I could watch right now?" means opening several apps in sequence at the exact moment you have sat down to watch — and titles saved months ago are effectively lost. Syncing those lists programmatically turns out not to be viable for anyone: there is no public API, no partner route, and the services' terms prohibit automation.

nextup takes the only sanctioned path left. You screenshot each service's saved list, upload the images in an explicitly declared append-only or full-update batch, and nextup extracts the titles by OCR/vision, matches them to TMDB metadata, and merges them into one deduplicated combined list — one row per title, a badge per service — with every change passing through your review before it is applied. The list never loses anything without asking you first.

The current release is single-user, mobile-first responsive web with federated sign-in. It supports Netflix, Max, Prime Video, Disney+, Apple TV+, Paramount+, Starz and Peacock, plus a separate rental-discovery waiting list and IMDb ratings. It is hosted on Azure, with a dated baseline estimate of ≈$11–14/month (cost assumptions), with no streaming credentials, no scraping, no scheduled jobs, and no telemetry. It's a personal, non-commercial project; it exists to test one premise: that a combined list is worth the price of feeding it by hand.

Status

Feature implementation is recorded in the generated task status, including the remaining owner-dependent acceptance work. Read the current release and recorded owner decisions and the requirement index before using older planning or review documents. Task completion is not a declaration that live quality, device checks or release acceptance are complete.

Quick start

Full instructions, including the offline test loop, are in docs/getting-started.md.

The install command is not run for you. Run it yourself.

Prerequisites

• Node 22 (see .nvmrc; nvm use)
• Docker — for the local SQL Server test container (mcr.microsoft.com/mssql/server:2022-latest) and Azurite blob emulator
• An Azure subscription only for deploying (not for local dev): Azure SQL Database Basic, Container Apps, Blob Storage, Azure OpenAI gpt-4.1, Azure AI Vision Read F0. See docs/architecture.md.

Install

npm ci
npx prisma generate --schema prisma/schema.prisma
cp .env.example .env   # then fill in the placeholders

Run

Follow getting-started section 5 to configure and start the API and the SPA in separate terminals. npm run dev starts only Vite; it does not start or authenticate the API.

Test

docker compose -f docker-compose.test.yml up -d   # mssql 2022 + Azurite
npm run db:test        # create nextup_test with the REQUIRED collation, then migrate
npm run test:unit      # Vitest — pure domain logic
npm run test:int       # Vitest — API surface against the mssql + Azurite containers
npm run test:web       # Vitest + Testing Library — component/screen states
npm run build --workspace apps/web
npm run test:e2e       # Playwright — the value loop and the irreversible paths

⚠ Do not create the test database any other way. It must be nextup_test and it must be created with COLLATE Latin1_General_100_BIN2 — which is exactly what npm run db:test does, and what CI does. A database created by any other route (a bare prisma migrate deploy against a new name, a manual CREATE DATABASE) gets the server default, SQL_Latin1_General_CP1_CI_AS, and then Prisma's create() joins its internal @generated_keys table variable — which takes the database default collation — against the BIN2 [id] column. Every insert fails with Msg 468.

The symptom points at the wrong layer: dozens of integration failures whose stacks all end in ownerData.ts, reading unmistakably as an application bug. Two people have now lost time to it. T-INV-018a is the only test that names the real cause — if it fails, fix the database, not the code:

SELECT name, collation_name FROM sys.databases;   -- must read Latin1_General_100_BIN2

See specs/testing.md §17.

Prepare the database environment and install Playwright browsers as described in getting-started before these commands. After the one-time dependency/browser downloads, Prisma generation and image pulls, the deterministic suites run offline — NFR-003 makes CI the only feedback loop, so the loop must not depend on a network. CI wires the same two containers as services (see .github/workflows/ci.yml and specs/testing.md §3.3a).

⚠ HTTPS is a functional dependency, not just a security control. navigator.clipboard does not exist on http://, so opening the dev server from a phone at http://<LAN-IP>:5173 shows no "Paste screenshot" button at all — and it looks like a missing feature, not a missing certificate. The desktop Ctrl/Cmd+V listener is unaffected. See docs/getting-started.md §6.

Project structure

nextup/
├── apps/
│   ├── api/            Node + Express API; in prod also serves the built SPA
│   └── web/            React + Vite single-page app
├── packages/
│   └── domain/         Shared, pure TypeScript domain (types, identity, rules)
├── infra/              Bicep infrastructure — Azure SQL, Blob, Container Apps (ghcr.io)
├── tests/              Cross-cutting unit / integration / e2e / infra tests + fixtures
├── docs/               BRD, PRD, architecture, backlog, roadmap, ADRs, diagrams
└── specs/              Implementation specs incl. the AC → named-test mapping

Documentation

Document	What's in it
Getting started	Clone → install → run the whole suite offline; the HTTPS/clipboard trap
Current release	Current scope, evidence boundaries, unresolved decisions and dated cost baseline
PRD	User stories and acceptance criteria; current scope is in section 11
Architecture	System design, the locked stack, and the cost model
Specs	Implementation detail; testing.md carries the AC → named-test mapping
Backlog	What to build, in order (the work order)
Roadmap	Sequencing and milestones
Parallel execution	How to run multiple coding agents at once, and the lane boundaries they must respect
ADRs	Why the load-bearing decisions are what they are
Runbooks	Operational procedures — start at the incident playbook, which routes by symptom. Includes the config checklist, rollback, the pre-authorised memory up-size and the re-used Vision account — one live grant that exists in no template
BRD	The (personal, non-commercial) business case
Review report	Historical pre-build review, not current task status

If you are a coding agent, start with .github/copilot-instructions.md.

Contributing

See CONTRIBUTING.md for setup, branching, commits, and what CI enforces. Work is driven by docs/backlog.md, in order; every acceptance criterion maps to a named test in specs/testing.md — that mapping is the definition of done.

Security

See SECURITY.md to report a vulnerability. nextup stores no streaming credentials, generates no SAS URLs, and sends screenshot bytes only to its two extraction endpoints.

License

MIT.