Search by words and meaning
Full text and meaning combined in one ranking, scoped to the document stores the product selects.
For requirement analysts
Your work is finding the rule, writing it so nobody can misread it, and showing later where it came from. KiwiPow turns your documents into citable sections, writes specs from them, and tells you when either side moves.
Your part in the loop
Word, PDF, Excel, Markdown, JSON and JSON Schema become trees of sections with stable ids. Tables stay tables, each sheet becomes a section.
One provable sentence per behaviour, edge cases written out, open questions named. Shape rules are checked as it is written.
Comment on any part, strike behaviours and answer questions, then submit the round as one. Revise with the planner or by hand in the same shape.
Departures from the spec land in the feedback queue. Accept one and the behaviour changes in a new revision, keeping its id.
At acceptance, act on the list of document sections the product has overtaken. KiwiPow changes no document itself.
Full text and meaning combined in one ranking, scoped to the document stores the product selects.
A document store is selected by every product it serves, never copied, so one law serves every product it governs.
A behaviour another spec already defines is cited, not repeated, and two drafts claiming the same behaviour are named and resolved.
Each comment is answered with what changed or a reason to disagree. A comment on text that has since changed is shown against the text it was left on.
Classified documents are absent to anyone not cleared for them, and the planner reads as the user it works for.
Search, section reads and document maps are offered over MCP, so any AI agent your team uses reads the same documents under the same access rules.
Import a document, write the ask in your own words, and read the citation on every behaviour the planner proposes.