Data sources

Rolling 14-day window of NYC Parks programming, with coordinates and times. No API key. Verified live on 2026-07-21. NOTE: dataset fudw-fgrp (which most tutorials use) is STALE — its newest row is 2019-12-28 — so we deliberately use w3wp-dpdi instead. Also note this is the same data as the nycgovparks.org RSS feed, which we do NOT fetch because that site's robots.txt disallows /xml/. The Open Data mirror carries no such restriction.

Street fairs, farmers markets, block parties, plaza events. No API key. Needs aggressive filtering: ~26k of ~32k rows in any 30-day window are youth/adult sports permits (ballfield bookings, not public events), and "Production Event" means film shoots. Has no coordinates, only a street description, so we link to a Google Maps text search rather than fake a pin.

GTFS-Realtime service alerts as JSON. No API key required (keys were dropped for subway/LIRR/MNR feeds). Verified live on 2026-07-21. Bus Time feeds DO still need a free key, so we use the keyless camsys all-alerts / subway-alerts endpoints only.

496 stations with coordinates and the routes serving each. Powers the travel-time estimator. Changes maybe twice a year, so weekly is generous.

Free weather forecast, no API key, no account, 10k calls/day for non-commercial use. We make 1 call per refresh. Requires CC-BY-4.0 attribution, which the status page provides.

Using previous data — the last successful fetch is still being served.

Parks, water fountains, bars. WEEKLY ONLY and never on the request path: during testing the public instance returned 504 repeatedly and one mirror was unreachable. The adapter retries across mirrors and, if all fail, keeps the previous committed snapshot and exits 0. Note OSM calisthenics coverage in NYC is very sparse (about one tagged node near Crown Heights), which is why curated workout spots are the primary source, not this.

The Skint and Brokelyn. Fetched once per day. Surfaced as LINK CARDS only — we do not parse their prose into structured events, because doing that reliably needs an LLM and would otherwise invent wrong times and prices. Both robots.txt files permit this. secretnyc.co was rejected: its robots.txt has an ambiguous "Disallow: /*/feed/".

Hand-researched recurring events, free museum hours, calisthenics spots and happy hours, stored as JSON in curated/. This is the only viable zero-cost source for trivia nights, run clubs and language exchanges — no free API provides them. Cannot break. Entries carry lastVerified dates and are auto-demoted after 90 days.

DISABLED — NOT A BUG. Eventbrite removed the public Event Search API in February 2020. There is no supported way to search events across the platform any more; only per-organization and per-venue lookups remain. The adapter can list events for an organization YOU control if you set ENABLE_EVENTBRITE=true plus a token and org ID. For discovering other people's events it is simply not available at any price short of their distribution-partner programme.

DISABLED — NOT A BUG. The open REST API is retired. The replacement GraphQL API needs an OAuth consumer, and creating one requires a PAID Meetup Pro subscription. That breaks the zero-cost, no-card constraint, so it ships off. The adapter is scaffolded so it is a small job to finish if you ever hold a Pro account.

DISABLED BY DEFAULT AND DELIBERATELY ISOLATED. tgtg-python is an unofficial client; automated access is contrary to Too Good To Go's terms and may get your account banned. It authenticates as you, via your real account email. It also breaks whenever TGTG changes their app. Nothing else depends on it: it runs in a separate Python process, and any failure is logged and ignored. Results are always tagged experimental and carry "Verify before going".