SignalScout Radar
A free public page of people asking which product to use or buy, sorted by
category. A product here is a physical thing: a gadget, a piece of gear, a
tool, a part or a supply. It lives at radar.buyerfinder.run.
Every hour a worker searches Reddit, and the newest threads of 10 forums from
different niches, for posts like "can anyone
recommend…", "what do you use…" or "is it worth buying…". A free text rule
sets most of them aside. One cheap model call per remaining post decides whether
it really asks for a product, which category the product is in, and what the
person wants. The page lists what survives: the category, the platform, a short
excerpt, the age and a link to the post. Never a name or a handle.
It is also a demonstration. The fetching and the model calls are
@signalscout/engine,
used on its own with no account, no monitor and no product profile. If you want
only the posts about your product, that is what
SignalScout does.
Limits it keeps
- A daily spending cap (
RADAR_DAILY_CAP_USD), providers and models
together. The worker checks the price of every call before it makes it, and
the day's spend is read from the database.
- An excerpt and a link. No username, no handle, no copied post, nothing
older than 30 days.
- Taken down on request. A removed post is marked, not deleted, so the next
run cannot put it back.
Running it
docker compose up -d # Radar's own Postgres on 127.0.0.1:5439
cp .env.example .env # fill in the keys; DATABASE_URL=postgres://radar:[email protected]:5439/radar
pnpm migrate
pnpm worker # a run every hour; add --once for one run
pnpm dev # the page on http://localhost:4321
pnpm test # needs the Postgres above
pnpm phrases [days] # requests kept per dollar, per phrase
A push to main tests, builds one image and deploys it by digest to the box
(.github/workflows/deploy.yml, scripts/deploy-remote.sh).
Backlog
Tickets share one id series with signalscout-open and signalscout-cloud,
and follow the rules in the open repository's
backlog/README.md.
