# How to choose a stack for a service platform

> How to pick a stack for platforms and dashboards, walked through a service desk on OpenThings: a UI shell, a data layer, domain APIs and an inbound phone channel.

- URL: https://www.rebels.software/guide/choose-stack-for-service-platform
- Published: 2026-11-07
- Author: Bart Szczepański
- Tags: Platforms, Dashboards, Tech stack, Field service, OpenThings

You need an internal platform or an operational dashboard and have to decide what to build it with. Pick the stack layer by layer: a web UI shell for screens and navigation, a data layer that matches how shared and durable the records must be, domain APIs that own the business rules, and channels that feed work in from outside. A service platform is a good worked example, because it needs all four.

## Why think in layers, not frameworks

Most stack debates start with a framework name. For platforms and dashboards that is the wrong end. The useful questions are simpler: who writes data, how long it must live, which rules move a record from one state to the next, and where new work comes from.

Our example is a manufacturer that services its own machines at customer sites. Customers call when a machine stops, an AI phone assistant takes the call, and a small team of technicians drives out from headquarters. The office runs everything from one web desk.

The shared foundation for this kind of work, at Rebels, is [OpenThings](https://openthings.net): a toolkit for building platforms. It already has the groundwork that usually eats the first months: infrastructure (on-prem or in your cloud), identity (tenants, single sign-on, scoped roles, an audit trail), and versioned contracts every module speaks. Modules sit on top. One of them is service: requests, technician plans, inspection protocols, and a phone assistant that books visits.

(Image: OpenThings homepage: every platform needs a foundation, with service and other modules stacked on that base)

## Step by step: assembling the stack

1. **Choose the UI stack for many screens and one frame.** The example uses Next.js with the App Router, React, TypeScript and Tailwind CSS, plus an icon library. The App Router gives you route groups, so every screen behind login shares one layout with a route guard. TypeScript keeps domain types such as a request, a visit or a machine in one place that screens and APIs both import. Tailwind reads design tokens from CSS variables, so brand colours are set once and used everywhere.
2. **Build the application shell once and configure it per project.** Login, a collapsible sidebar that becomes a drawer on small screens, a sticky header with breadcrumbs, a user menu, notifications and a few layout variants all live in a reusable shell. Name, logo, colours, languages and the navigation list sit in one config file. A new platform starts as a copy of the shell with that file filled in, so the first day already has a working app to log into.
3. **Match the data layer to who writes the data.** A dashboard shell can run on typed TypeScript modules for reference data and browser storage for a user's own state, such as a collapsed sidebar. A service desk cannot. The office, the technicians and the phone assistant all write to the same requests, and those records must survive a restart. The example adds SQLite through the database module built into Node.js, behind a thin server-only data layer, with tables for requests, visits, calls, call events and settings. It is one file with no extra service to run. When you outgrow it, the data layer is the only part that has to change.
4. **Model the domain before you add screens.** Write down the request lifecycle first. In the example a request opens when a call starts and ends as new (callback needed), scheduled, in progress, resolved by phone, done, or dropped when the call ended without enough information. Each request holds the machine and serial number, the issue, the checklist steps tried, the linked visit, the transcript, a summary and an event history. Static reference data, such as the machine catalogue, per-issue checklists built from each model's features, and the technician list, can stay in typed modules because the office rarely edits it.
5. **Put business rules behind domain APIs.** Route handlers in the same Next.js app expose requests, visits, technicians, settings and the dashboard summary. Visit planning lives here, not in the browser: slots are ranked by day, then by distance from headquarters or from the nearest visit already booked that day, within a maximum distance set in settings. Every screen and channel then gets the same answer.
6. **Add the phone channel as an integration, not as the core.** The AI phone assistant gets a narrow set of HTTP actions: call started, find the machine, get the checklist, check for an open request by serial number, get available slots, book a visit, call ended. It never touches the database directly, and the endpoints are protected with a token. Every action is logged as a call event, so the request detail page can show a live view of the call by polling a small endpoint every second or two. At the end the request moves to its final state with the transcript attached.
7. **Build the screens on top of the model.** The dashboard shows today's numbers, recent calls with the active one on top, and who is on the road. Request detail shows the checklist, transcript, visit, history and notes. A calendar with a route map, technician profiles, a product catalogue and a settings page complete the desk.

(Image: Service desk dashboard with sidebar navigation, an ACME Corp wordmark, KPI cards and a 14-day requests chart)

## Two layers you can see in the example

The shell alone is a dashboard stack: UI framework, typed modules and config. It is enough for prototypes, read-only views and branded demos. The service platform adds operational data, a request lifecycle, calendar and routing rules, and an inbound channel. That is the full example of a stack: UI shell, plus operational data, plus channels.

(Image: OpenThings modules on the same foundation, including service)

## When this stack fits, and when to reach for more

This kind of web-app stack fits custom operations software, multi-screen desks for one team, and branded client portals where one codebase serves the UI and its API. It keeps the moving parts few: one app, one database file, one deploy.

Reach for something heavier when several systems must share the same rules, when the load or the number of concurrent writers outgrows a single database file, or when an ERP or CRM already owns the records. Then the platform becomes a client of that system or of a separate backend service, and the web app keeps the screens and the shell.

## Common mistakes

- **Choosing the framework first and the data model last.** The screens end up unable to show a dropped call or a visit booked by the assistant.
- **Keeping shared records in browser storage.** It works in a demo and breaks the moment two people use the desk.
- **Hardcoding colours, labels and navigation.** Every new project then copies the chrome and drifts. Keep them in config and dictionaries.
- **Letting a channel own the data.** The assistant is one way in, next to the office and a web form. Requests, visits and history live in the platform, where people review and correct them.
- **Calling the database from route handlers everywhere.** Put queries in one data layer, so swapping SQLite for a larger database later is one change.

## Next step

Write down, on one page, who writes each kind of record, how long it must live, and which channels create new work. That page tells you which layers your platform needs. If you are also deciding what to automate first, see [how to choose the first process for an AI agent](/guide/choose-first-process-for-ai-agent). To see the foundation these layers sit on, start at [OpenThings](https://openthings.net).

If you want to talk through the stack for your own platform, book a call on this page to start with a Discovery workshop, or begin from the [Rebels Software homepage](/).

We start with your biggest problem and the data behind it. [Book a free call](https://calendly.com/bartlomiej-jacyno/intro)
