What it is
The Anthropic API is the developer interface to Claude, Anthropic's family of large language models. Applications send a conversation through the Messages API and get Claude's reply back. It accepts text, images and PDFs, so it can read documents as well as answer questions. Developers authenticate with an API key and pay per token of input and output.
Beyond plain chat it supports tool use, where Claude calls functions the developer defines, structured output that fits a set schema, and prompt caching to cut the cost of repeated context. The same Claude models are also offered through cloud platforms such as Amazon Bedrock. Anthropic publishes several model sizes, trading speed and cost against capability.
How I use it
- Lend Pilot pulls numbers out of pay stubs, NOAs and statements with Claude so nobody retypes them.
- Claude extracts structured data from uploaded borrower documents, and a human checks the output.
- Every dollar figure, qualification and stress test stays in deterministic code. The model never does the arithmetic.
- Rental Pilot's Radar assistant and the website assistants run on Claude through Amazon Bedrock rather than this API directly.
The Anthropic API is the right tool when a system has to read messy, human documents and turn them into fields. I reach for it for exactly that, and not for arithmetic: the model reads, a person checks, and qualification and stress tests run in ordinary code so the result holds up to a regulator.
Questions
Can AI be trusted with mortgage numbers?
In Lend Pilot the model only extracts figures from documents, and a person checks them. Every calculation, from qualification to the stress test, runs in deterministic code.
Does Shawn Mervin build with Claude?
Yes. Lend Pilot uses the Anthropic API for document extraction, and Radar in Rental Pilot plus the assistants on several websites run on Claude through Amazon Bedrock.