Carbon (carbon.ai) was a data connector service for applications built on large language models, made by JCDT, doing business as Carbon. The carbon.ai site no longer loads; this description comes from the Internet Archive copy of its home page from 2 January 2025, so the service is listed as closed.
Carbon offered one API with more than 25 pre-built connectors. They ingested unstructured data, such as files from Google Drive, Notion, Dropbox or OneDrive and the text of websites, and loaded it into a vector database. A vector database stores text as embeddings, lists of numbers that capture meaning, so an AI app can find the passages closest to a question; this method is called retrieval-augmented generation (RAG). Carbon parsed more than 20 file formats into plain text and Markdown, split the text into chunks, created embeddings, synced each source on a custom schedule and offered hybrid semantic and keyword search. It also handled OAuth logins to third-party services.
It was a hosted API billed per character synced, with a monthly Professional plan as an option; a self-managed deployment was available on custom contracts.