Table-Scout: what's open now, and where can we book?
A restaurant finder for Stavanger and Sandnes that answers plain questions like "what's open now?" or "where can four of us eat at seven?". It builds its own venue list from open map data and the official business registry, and uses Claude to pick and explain the best matches in the asker's own language. A private tool, built and run by Agentas.
- Status
- Live since August 2026, private use
- Coverage
- About 950 venues in Stavanger and Sandnes
- Stack
- Node.js, TypeScript, SQLite, Claude
- Source
- Private — walkthrough on request
What it does
Its own list of venues
Restaurants, cafés, bars and fast food from OpenStreetMap, matched against the Brønnøysund business registry to fill the gaps. The registry roughly tripled the list, to about 950 venues.
Open now, from real opening hours
Opening hours come from the map data and from each venue's own website, where Claude reads the page text and turns it into structured hours. "Open now" is then calculated from those hours.
Can we book?
The tool detects which booking system each venue uses, such as Superb, Gastroplanner or easyTable. Where a provider publishes a status like "fully booked", it shows it; otherwise it gives a one-click booking link. It never makes up free time slots.
Ask in plain language
Ask for nachos, coffee or a table for four, in Norwegian or English. Common cravings are mapped to cuisines in code, a shortlist of up to 40 venues goes to Claude, and the answer comes back as ranked picks with a short reason for each.
Ready for voice
Every answer also comes as one spoken sentence in the question's language, so a Home Assistant voice assistant can read the recommendation aloud.
A dashboard with a person in the loop
Tabs for open now, browse and ask, with hours for each day of the week and the distance from a chosen starting point. Anyone using it can correct the hours, mark a venue closed or add its website right on the card.
Relevant if you're looking for…
Table-Scout is a private tool without public access; these are the needs it covers, and we build similar tools for businesses.
- Restaurant opening hours: they come from map data and from each venue's own website, where Claude turns the page text into structured hours, and “open now” is calculated from them.
- Booking a table: the tool detects which booking system each venue uses, shows “fully booked” only where the provider publishes it, and otherwise gives a direct booking link without inventing free slots.
- Restaurant recommendations from AI: ask in English or Norwegian for nachos, coffee or a table for four, and Claude ranks a shortlist of up to 40 venues with a short reason for each pick.
- AI search over local data for your business: we build tools like this from open data and official registries, with AI to pick and code to check the answer.
We build this for clients: AI and LLM integration into your systems · Data and reporting platform
How it works
- 01
Collect
Map data and the business registry, refreshed every week.
- 02
Match
One list of venues, with duplicates between the sources set aside.
- 03
Enrich
Address, opening hours and booking system for each venue.
- 04
Answer
Claude ranks a shortlist and explains the picks.
How it's built
Table-Scout is one Node.js and TypeScript service with a SQLite database. It runs as a single Docker container with a read-only file system, all Linux capabilities dropped and a non-root user, and every push to the main branch is type-checked and unit-tested before it deploys.
The data rules came first. OpenStreetMap data, under its ODbL licence, lives in its own table apart from what we collect ourselves, and Google Maps is never used as a source. Claude runs with every tool switched off and picks only from a numbered shortlist; any pick that isn't on the list is dropped in code.
Engineering highlights
- Every tool switched off for the AI model, enforced in two separate layers
- Booking answers use only what providers publish openly; no invented time slots
- Map data kept apart from our own data, so the licence boundary stays clean
- Venues, hours and booking systems refreshed automatically every week
- Read-only, non-root container; type checks and unit tests gate every deploy
Questions
Can I use Table-Scout myself?
No, it is a private tool built and run by Agentas for our own use. The source is private, and we give a walkthrough on request.
Where does the restaurant data come from?
From OpenStreetMap, matched against the Brønnøysund business registry to fill the gaps, and refreshed automatically every week. Google Maps is never used as a source.
Does it work with a voice assistant?
Yes. Every answer also comes as one spoken sentence in the question's language, so a Home Assistant voice assistant can read the recommendation aloud.
Want something like this?
Table-Scout shows how we work with data that comes with rules: use the sources you are allowed to use, keep licensed data apart, let AI read and rank, and let code check the answer.
Get in touch