Ask the corpus a question in plain language before you know the right search term for it, running entirely on your own data.
Type a plain-language question against a saved search's scope and get back the documents that semantically match, a way to find the vocabulary and names worth searching for before writing the real, deterministic search.
Explore always runs against a saved search you pick first, so it works a defined universe rather than the whole case.
The interface shows its own engine status (local, CPU), since, unlike cloud-based AI review tools, it doesn't call out to a cloud model.
Extract runs up to ten standing questions across every document in a chosen scope in one pass, instead of one conversational query at a time.