get_text_fragments Leave feedback

get_text_fragments returns the passages that match, each with a few words of context, grouped per document — the call an agent makes to quote a clause instead of reading a whole file. It takes the same query language as search. Example prompt: “What do the contracts say about the termination penalty? Quote it.”

Tool description (as the AI agent sees it):

Returns the passages of indexed documents that match a query — each match with the words around it — so the user can read what a document says without opening it. Call this when the user wants to see the matching text, quote a clause, or answer a question from their documents (‘what does the lease say about termination?’), usually after search has shown which files match. Takes the same query syntax as search. Works on documents indexed by this server (text is stored in the index). Returns plain text: a header line with the document count, then per document a ‘=== ( occurrence(s)) ===’ line followed by up to maxFragmentsPerDocument fragments, each prefixed with the field it came from. In evaluation mode at most 10 documents are considered. On failure the response starts with ‘Fragment extraction failed for’ followed by the reason, and isError is set.

{
  "name": "get_text_fragments",
  "arguments": {
    "query": "\"termination penalty\"",
    "indexName": "contracts"
  }
}

Plain text. Captured from server 26.10.1 (evaluation note trimmed):

Query '"termination penalty"' in index 'contracts': 2 document(s) match.

=== contracts/acme-master-services-agreement.pdf (1 occurrence(s)) ===
  [content] …terminates before the end of the Term, a termination penalty of
three monthly fees applies, payable within 30…

=== contracts/globex-consulting-agreement.docx (1 occurrence(s)) ===
  [content] …at any time with 30 days notice.
Early termination penalty: the Client pays 50 percent of the fees…

• termsBefore / termsAfter (default 8 each) set how many words of context surround a match — raise them to 20 for a full clause.
• maxFragmentsPerDocument (default 3) and maxDocuments (default 5) cap the size of the answer.
• Each fragment is prefixed with the field it came from — usually [content]. A match in a file name or a file date appears under that field instead.
• A match inside an email attachment shows as === mail/message.eml → attachment.docx (…) ===.

Context is counted in words, not sentences, so a fragment can start mid-sentence. On failure the response starts with Fragment extraction failed for and isError is set.

• “Quote the termination penalty clause from each contract.”
• “Show me where the datasheet mentions the operating temperature.”
• “What does the quotation in my mail say about the price increase?”

See it used end-to-end: Quote matching passages, not whole files.