PDF question answering

ChatPDF answers natural language questions about the content of an uploaded PDF. A Beam agent sends a question when a new document arrives, such as a contract or invoice, and reads back the answer ChatPDF returns from the file's text. It applies an approved rule to decide what the answer means, for example whether a contract clause matches a required term, and writes that result into a tracking sheet or record. When the answer is ambiguous or the document format cannot be parsed cleanly, the agent flags the file for a person to check by hand.

Structured data extraction from PDF text

ChatPDF can pull specific fields out of a PDF's text, such as a total, a date, or a name, when asked directly. A Beam agent triggers this extraction after a document is uploaded, reads the returned values, and writes them into the fields of a linked record like an invoice tracker or contract log. It applies the approved rule for formatting or validating each value before saving it. If a field is missing, unreadable, or the extracted value looks wrong, the agent routes the document to a person instead of guessing.

Session based document chat

ChatPDF keeps a chat session open against one document, so follow up questions can reference earlier answers within the same file. A Beam agent uses this to ask a sequence of checks against a single contract or report, reading each reply before deciding the next question based on the rule it was given. It writes the final set of answers into a record once the session completes. When a reply contradicts an earlier one or the document runs out of relevant content, the agent hands the session to a person to finish.