Liminary is a memory layer for people who read for a living. Articles, reports, PDFs, AI chats and videos are saved with one click, and the collection then works for you: Magnet surfaces relevant material during meetings unprompted, claims can be fact-checked against your own saved sources, gap detection flags what a document is missing, watchlists monitor topics for anything new, and Gmail and Notion integrations draft using the saved context. Every insight carries its source.
Source traceability is the design decision that matters. A retrieval tool that produces a confident summary with no provenance is unusable for consultants, analysts and investors – the audiences named here – because the whole job is being able to say where a claim came from. Fact-checking against your own library rather than the open web is the same instinct applied well: your saved sources are ones you have already judged, which is a materially better corpus than whatever ranks highest today.
The headline number needs reading carefully. ‘3.4X more accurate than ChatGPT at retrieving the right answer across your internal and external sources’ compares a purpose-built retrieval system against a general assistant on a task the assistant was not built for, which is a comparison designed to be won – it says little about how Liminary performs against other retrieval tools, which is the actual decision. The Magnet feature also implies listening in meetings, so check what is captured and where it is stored before switching it on. No rates are published on the landing page.
