How it works

Enter your products once. We watch FDA for them.

Not a search engine you have to remember to use. Not a news feed you have to filter. A standing instrument that already knows what you make.

1 — One identifier builds your product profile

Give us a product code, a K-number, or just what the device is called. FDA’s own classification database supplies the rest — device class, CFR regulation, submission pathway, review panel, the implantable and life-sustaining flags. You confirm it in one click.

We derive rather than askbecause a thirty-field form costs you an afternoon and produces typos in exactly the identifiers that matching depends on. Anything FDA’s record doesn’t say — where the product is in its lifecycle, who your contract manufacturer is — we ask for, because those two facts are the ones that change what’s urgent for you.

What one product code returns
IRT
→ Pad, Heating, Powered
→ Class II
→ 21 CFR 890.5740
→ 510(k) Exempt
→ Physical Medicine panel
→ Not implantable · Not life-sustaining

Live from FDA’s classification database — try it on the homepage.

2 — Relevance is a join, not a guess

FDA keys its own data on identifiers: 510(k) clearances carry product codes, classification changes carry CFR citations, warning letters carry firm names. Your product profile holds the same identifiers. Matching is therefore a database join — mechanical, auditable, and incapable of hallucinating.

The “why this affects you” line on every match is not generated prose. It is the join key, shown: “Cites 21 CFR 876 — your regulation.”

Where FDA publishes no key — recalls, notoriously, carry none — we do not fall back to a semantic guess, because “your device is recalled”said wrongly is the worst sentence this product could emit. Instead, the reviewer who already reads every Class I recall supplies the missing product code by hand, and that one keystroke becomes an exact match for every subscriber at once. Matches are labelled with where their key came from — FDA’s record or our reviewer — and you can always tell which.

3 — A person reviews everything before you see it

The model reads FDA. It drafts summaries and extracts identifiers — work it is good at and where being wrong is visible. It is never allowed to assign urgency or write a recommendation. Those two fields are written by a regulatory professional, for every single item, and the database physically refuses to publish an item without them. This is not a review step we hope to automate away; it is the product.

The arithmetic that makes this possible: after filtering the firehose (roughly 250 routine Class II recalls and 200 clearances a month go to the searchable corpus, not the queue), what remains is ~18 material items a week. A person can genuinely review that. A person could not review 400, which is why volume discipline is a safety property here, not a growth problem.

4 — The digest is the product

One email a week: what changed, why it matters to your products, what to do about it — urgency first, never chronology. A Critical item on your watchlist goes out the day it is approved, not on Monday.

And on the weeks nothing touched you, the email says so with evidence: “We checked 92 FDA documents across 4 sources. None touched your products.” That sentence is the entire promise — you get to stop checking FDA yourself, because someone provably did.

See it against your own product.

One identifier. Ten seconds. Your regulatory landscape.

Start monitoring