The 44B API

One authenticated call searches every corpus at once — the document corpus, released models, the benchmark catalogue, the provider roster, and the incident record — and a second turns that same blend into an answer where every claim carries a citation you can open.

44B is an independent reference implementation of New York General Business Law Article 44-B. It is not a government service and carries no official status. The corpus is refreshed nightly, so most of what it holds postdates the training cut-off of the model you are probably asking — which is the point of querying a registry rather than a model.

What it is not: there is no write access — nothing in the API files, submits, or edits anything; no personal data — the corpus is published documents, public filings, and the public incident record, and filings Article 44-B makes confidential are never returned; and no legal advice — it returns records and citations, not conclusions about your obligations.

3,486,939

Documents

3,485,585

Papers

586

Models

663

Benchmarks

2,568

Incidents

Providers

Live counts, read from the same tables the API serves and refreshed hourly on this page. Every figure is reachable without a key — the website is the free tier.

Your first call

Three steps, about two minutes, no client library.

Create an account

Mint a key

On /account, create a key on Pro or Power. It is shown once — 44B stores only its hash — and sent as Authorization: Bearer sk44b_….

Call the API

One GET, no body, no SDK. Everything below runs against the live corpus the moment your key exists.

The call

curl

curl -H "Authorization: Bearer sk44b_…" \
  "https://44b.nysgpt.com/api/v1/search?q=frontier%20model"

JavaScript and Python versions of every example are on the reference.

What comes back

{
  "query": "frontier model",
  "page": 1,
  "pageSize": 15,
  "total": 73,
  "totals": {
    "library": 2081,
    "models": 0,
    "benchmarks": 2,
    "incidents": 11
  },
  "results": [
    {
      "id": "2607.08700",
      "corpus": "library",
      "title": "Do You Need a Frontier Model as a Citation Verifier? …",
      "href": "/library/2607.08700",
      "date": "2026-07-09",
      "score": 600
    }
  ]
}

totals reports every corpus the query touched, so one call tells you where the evidence actually is. Full field-by-field breakdown in the reference.

Recipes

Five real questions and the exact call that answers each one. Every search here was run against the live API before this page shipped.

Question

Which incidents on the record involve AI giving medical advice?

Returned · 5 incidents. Top of the list: “Alleged Harmful Health Outcomes Following Reported Use of Purported Chatbot Advice” and “Microsoft Copilot Allegedly Provides Unsafe Medical Advice”.

curl

curl -H "Authorization: Bearer sk44b_…" \
  "https://44b.nysgpt.com/api/v1/search?q=medical+advice&corpus=incidents"

Question

What does the literature say about models gaming their own evaluations?

Returned · 342 matching documents in the corpus, 60 returned in the blended set — title and abstract are both searched.

curl

curl -H "Authorization: Bearer sk44b_…" \
  "https://44b.nysgpt.com/api/v1/search?q=reward+hacking&corpus=library"

Question

Find a lab’s published safety framework, and the research around it.

Returned · Anthropic’s Responsible Scaling Policy at rank 1, followed by 59 further records — the lab document and the literature citing it, in one call.

curl

curl -H "Authorization: Bearer sk44b_…" \
  "https://44b.nysgpt.com/api/v1/search?q=responsible+scaling+policy"

Question

Which open-weight models are on record, and who builds them?

Returned · 32 models, each resolving to a /models page carrying benchmarks, price, throughput, licence, and the providers serving it.

curl

curl -H "Authorization: Bearer sk44b_…" \
  "https://44b.nysgpt.com/api/v1/search?q=llama&corpus=models"

Question

Just answer the question, with citations I can check.

Returns · Prose grounded only in records 44B holds. Every claim carries a bracketed citation indexing into sources, and every source has an href that resolves to a real page. Costs one answer against your monthly allowance; identical questions are cached for seven days and cost nothing.

curl · POST

curl -X POST "https://44b.nysgpt.com/api/v1/answer" \
  -H "Authorization: Bearer sk44b_…" \
  -H "Content-Type: application/json" \
  -d '{"question": "What does the incident record show about algorithmic decision systems used in benefits?"}'

Three ways in

The same corpus and the same key, reached three different ways.

REST

Two endpoints. GET /api/v1/search returns the same blend the website searches; POST /api/v1/answer turns that blend into a cited answer. JSON in, JSON out, nothing to install.

Read the reference →

MCP

Bring your own model. 44B speaks the Model Context Protocol over Streamable HTTP with nine read-only tools over the live database, authenticated with the same key and metered on the same counters.

Connect a model →

/chat

The 44B agent, on the site rather than the API. It answers by querying the database in front of you, shows each tool call as it runs, and links the page every claim came from.

Open the agent →

Plans

The website is free and unmetered. What a subscription buys is programmatic access.

Free

The whole website, open to everyone.

Searches / mo: —
Answers / mo: —
API keys: 0

Pro

$49/mo

Programmatic access for researchers and journalists.

Searches / mo: 10,000
Answers / mo: 200
API keys: 3

Power

$99/mo

Volume access for insurers, analysts, and teams.

Searches / mo: 100,000
Answers / mo: 2,000
API keys: 3

Start free. Add the API when your questions outgrow the search box.

Create your account View plans