Data infrastructure for software that pays its own way.

Cabrini is a market data service designed from the ground up for AI agents — the fastest-growing consumers of financial data, and the ones the industry never planned for.

The problem we saw

Every market data provider assumes a human is on the other end. Someone to fill out the signup form, verify an email, enter a credit card, manage API keys, and remember to cancel the subscription. That model made sense for twenty years. It makes no sense for an autonomous agent that needs one day of price history for one stock, right now, in the middle of a task.

So agents either go without data, or their operators pre-provision accounts and keys for every service the agent might someday touch. Both options are bad. One starves the agent; the other buries the human in setup work.

What we built instead

Cabrini removes the account entirely. When an agent requests data, the response quotes an exact price. The agent signs a USDC micropayment on the Base network using the open x402 standard and retries the request with proof of payment. Data comes back. The whole exchange takes one round trip and costs a fraction of a cent.

There is nothing to sign up for, nothing to store, and nothing to leak. No API key can be stolen because no API key exists.

The dataset

Price data
Every US equity — including delisted symbols — from September 2003 to the present. Eight intraday timeframes (3-minute to 4-hour) plus daily bars with VWAP.
Fundamentals
Quarterly financials normalized from SEC EDGAR XBRL filings — revenue, net income, EPS, margins, cash, debt.
Filings
10-Ks, 10-Qs, 8-Ks, proxies, and ownership filings — with optional full-text section extraction (risk factors, MD&A) so agents skip parsing 300-page documents.
Insiders
SEC Form 4 transactions: who bought or sold, role, shares, price, and value.
Breadth
7,000 to 16,000+ tickers per trading day
Quality
Gap-checked, normalized to a single schema, refreshed daily

How agents find us

Cabrini publishes machine-readable descriptions of itself over every major discovery standard: an MCP server for tool-calling models, an A2A agent card, an OpenAPI spec, and llms.txt. An agent that has never heard of Cabrini can discover the service, read the price list, and complete a purchase without a human touching anything. The full technical picture lives in the agent integration guide.

Where this is going

We think most API traffic in ten years will come from software acting on someone's behalf, paying per use, with no standing relationship to the services it consumes. Market data is our starting point because it's what agents ask for most — but the model is the product: open protocols, transparent per-query pricing, and zero onboarding.

Payments settle in USDC on Base mainnet through the x402 protocol.