FDA device record indexed through Oct 9, 2026 · refreshed weekly

Know what FDA already cleared.

Ask about predicates, testing, standards and recalls. Get an answer where every precise fact opens to the page of the 510(k), De Novo or guidance it came from.

US FDA devices: 510(k), De Novo, guidance, 21 CFR

Free for three cited answers a month. No card.

No uncited claimsBuilt for 510(k), De Novo and PMA teamsFree Predicate Finder, no accountNow onboarding design partners: 6 places left

A device question, answered with the record

Bearnier AI reads the clearances, De Novo decisions and guidance behind your question, writes an answer, and ties every precise fact to the page it came from. Click a citation to see its source.

What predicates have recent AI triage devices cited, and what testing did FDA expect?

Cited 7 sources · counted 88 FDA records
  1. Mapping the question to product code QAS
  2. Reading 88 records under 21 CFR 892.2080
  3. Opening De Novo decision summary DEN170073
  4. Opening 510(k) summaries K261085, K260729 and K261317
  5. Extracting predicates and standalone performance

AI triage devices are cleared under product code QAS, a Class II device regulated at 21 CFR 892.2080 , with 88 510(k) records in openFDA . The category was created by De Novo DEN170073, Viz.ai’s ContaCT .

Recent submitters cite an earlier cleared triage device, not the De Novo. K261085, iSchemaview’s Rapid VO, cleared Jul 13, 2026 , flags suspected large vessel occlusion on adult head CTA and names the company’s own Rapid CTA 360, K251151 . K260729, GuideAI Health’s VAOT, cleared Jun 18, 2026 , names Aidoc’s BriefCase LVO Triage, K220709 , although VAOT flags occlusions in the aorta and the iliac, femoral and popliteal arteries . Aidoc’s own Special 510(k), K261317, also names K220709 .

FDA accepted standalone performance on a few hundred multi-site cases, reported as sensitivity and specificity with 95% confidence intervals, plus time to notification:

  1. K261085: 314 CTA cases from 33 clinical sites (99.0% US), 282 evaluable with 113 positive and 169 negative . Sensitivity 0.956 (95% CI 0.901–0.981), specificity 0.905 (95% CI 0.852–0.941), mean time to notification 3.2 to 3.3 minutes
  2. K260729: 317 cases from US sites, 165 positive and 152 negative . Sensitivity 0.915 (95% CI 0.862–0.953), specificity 0.901 (95% CI 0.842–0.944), time to notification under 5 minutes

In both, ground truth came from three board-certified radiologists by majority vote .

13 citations
Open
Open
Open
Open
Open
Open
Open
Open
Open
Open
Open
Open
Open

Used by device RA teams, consultants and founders

“We used to spend a week building the predicate table for a new 510(k). Bearnier gave us the table and the pages behind every row before lunch.”
Director of Regulatory Affairs
Cardiovascular device company
“What sold me is that every claim opens to the summary page. I can hand it to a reviewer without redoing the research.”
Senior RA Specialist
Imaging AI startup
“It found two predicates in our product code that our consultant had missed, with the testing each one ran.”
VP Quality and Regulatory
Infusion therapy manufacturer
“I ask it the questions I used to save for our monthly call with the consultant.”
Co-founder and CEO
Seed-stage wearables company
“The standards answers are the part I did not expect. It told me which edition FDA recognizes and how three competitors declared conformity.”
Systems Engineer
Surgical robotics company
“Our first Pre-Sub briefing package went from three weeks to four days.”
Regulatory Consultant
15 years in devices

Built for how devices get cleared

Device submissions argue from precedent, standards and field history. Bearnier AI reads all three.

Precedent

Every clearance in your product code, compared on intended use, technology and the testing FDA accepted. Including the scanned summaries keyword search misses.

QAS· 88 records
DEN170073· created the category
K261085· cited 5 times in answer

Standards

Which ISO and IEC editions FDA recognizes, how they apply to your form factor, and how comparable devices declared conformity.

IEC 60601-2-24· infusion pumps
ISO 81060-2· NIBP accuracy
IEC 62304· declared in 9 of 10

Signals

Recalls, adverse events and warning letters for devices like yours, so you see what FDA worries about before your reviewer asks.

FRN recalls· software, alarms
MAUDE· failure modes by code
Warning letters· CAPA, complaints

From question to submission section

  1. 1

    Ask

    Type the question the way you would ask a senior colleague. Product codes, K-numbers and plain descriptions all work.

  2. 2

    Check

    Open any citation to the exact page. Bearnier AI says when the record does not support an answer instead of guessing.

  3. 3

    Draft

    Turn the answer into an SE comparison, testing summary or Pre-Sub question set, exported to Word (coming soon).

  4. 4

    Watch

    Follow product codes, and soon competitors. Get an email when FDA clears, recalls or publishes guidance that touches them.

What Bearnier AI reads

The US FDA device record, indexed as one corpus and refreshed weekly.

Dataset
What it answers
Status
510(k) notifications and summaries
Predicates, intended use and the testing FDA accepted
Live
De Novo decision summaries
How new device categories and their special controls were set
Live
Product classification
Product codes, regulation numbers and device class
Live
CDRH guidance
What FDA expects to see in a submission
Live
21 CFR 800–898
Classification regulations and device requirements
Live
PMA approvals and SSEDs
Clinical evidence behind Class III approvals
Q1 2027
Recognized consensus standards
Which standard editions FDA recognizes
Q1 2027
Recalls
Field problems and root causes by product code
Q1 2027
See every dataset and the roadmap

No uncited claims. Measured, then published.

Bearnier AI’s model only states a precise fact when it can show you the page. To prove it, we grade every release against 200 device questions written and answered by regulatory professionals, and publish the scores.

January 2027
First published scores
200
Expert-written device questions
3
Scores per release: answer accuracy, citation validity, correct refusals

Get the question set when we publish it (first results January 2027).

Become a design partner

We are working closely with a small group of device teams with a submission in the next six months. 6 places left.

You get
Team plan at half price for 12 months and a direct line to our team
You give
Weekly use on a real submission and a 30-minute call every two weeks
We build
The answers, drafts and alerts your submission needs first

Simple pricing for device teams

Start free. Upgrade when Bearnier AI is part of your submission work.

Free
Try it on a real device question
$0
No card
Start free
  • 3 cited answers a month
  • US FDA device records
  • Predicate Finder with CSV
  • The Clearance Log
Pro
One RA professional or consultant
$250/ month
Or $2,500 a year, 2 months free
Subscribe to Pro
  • Unlimited cited answers, under fair use
  • Saved threads
  • Deep research runs (coming soon)
  • Word export with clickable citations (coming soon)
TeamBest for teams
A device RA and quality team
$1,000/ month
5 seats
Subscribe to Team
  • Everything in Pro for 5 seats
  • Onboarding call
  • Private library of your submissions and FDA letters (coming soon)
  • Alerts by product code and competitor (coming soon)
Enterprise
Larger portfolios with security review
Custom
Book a demo
  • SSO and audit logs
  • EU MDR sources (2027)
  • Custom DPA
  • SOC 2 (planned 2027)

Secure checkout by Stripe, with Link for one-click payment. Cards, and invoices for annual plans.

Questions device teams ask

Something else? Book a demo

What is Bearnier AI?

Bearnier AI is regulatory intelligence software for medical device teams. You ask a question in plain language and get an answer where every claim links to the exact page of the FDA record it came from.

How is Bearnier AI different from ChatGPT or Claude?

General assistants answer from training data and the open web, which covers famous devices and misses most of the clearance record. Bearnier AI searches an index of FDA device documents, including scanned PDFs, and only states what it can cite to a page.

Can Bearnier AI find a predicate device for my 510(k)?

Yes. Describe your device or enter its product code and Bearnier AI lists cleared devices with the same intended use and technology, the predicates those devices cited, and the testing in their summaries.

Which regulators does Bearnier AI cover?

The US FDA device record today. Recognized standards, recalls and PMA records arrive in early 2027, and EU MDR sources follow in 2027.

Does Bearnier AI hallucinate?

Bearnier AI is built not to. Its model only states a precise fact when it can cite the source page, and tells you when the record does not support an answer instead of guessing.

Is my submission data private?

Team library documents are visible only to your workspace, are never used to train models, and can be deleted at any time.

How much does Bearnier AI cost?

Free for three cited answers a month; Pro $250/month or $2,500/year; Team $1,000/month or $10,000/year for five seats; Enterprise custom.

Who builds Bearnier AI?

A team with deep regulatory and AI engineering expertise, working alongside device regulatory professionals who grade every answer and write our guides.

The Clearance Log

Every Monday: notable clearances, new device guidance and recalls, filtered to the product codes you follow. Five minutes, every item cited.

Sample issue · Monday, Oct 12, 2026
This week in DPS
K254255S-Patch CardioAI cleared, Sep 25, 20261

Know what FDA already cleared.