Claude Code, Cursor and Cline find code with grep, not embeddings: exact matches are cheap, checkable and never stale. But grep stops at one laptop's files. StackGrep is the same loop over everything else your agent needs: tickets, contracts, logs, docs and buckets, indexed so a search takes milliseconds instead of reading every byte.
# what your agent already does in a repo
grep -rn "INV-1042" .
# the same thing over your company's data
search_collection({ collection: "billing", query: "INV-1042" })
search_collection({ collection: "billing", query: "INV-10[0-9]{2}", regex: true })
count_collection({ collection: "billing", query: "chargeback" })search_collection(query: "INV-1042")search_collection(query: "db-old\.internal", regex: true)count_collection(query: "chargeback")Because exact matches are reliable: the agent searches, reads what it found, and searches again, and nothing goes stale when files change. The teams behind Claude Code and Cline have both said they chose search over embeddings for this reason.
Plain grep is: it reads everything. StackGrep keeps an index of every 4-byte sequence in your text, so a search only reads the documents that can match. That's how it answers over millions of documents in milliseconds.
Results come back as the matching line with its document id, not whole files, and count_collection answers how many without returning them. The agent reads a whole document only when it asks for one.
StackGrep Engine is onboarding teams from the waitlist. Bring your documents or your bucket.