Cortex
The research workspace where every AI answer can show its work — cited to the passage it came from, and costed before it is written.
The interface, up close.
The AI workspace where every answer can show its work.
Cortex is a research workspace for teams who have to stand behind what AI tells them. Most AI tools hand you an answer and ask you to trust it. Cortex works the other way: it answers from your team’s own documents — research, customer interviews, product specs, legal and policy notes — and footnotes every claim to the exact passage it came from, so anyone can open the source, read it in context, and judge for themselves whether the answer holds up.
Your documents live in knowledge bases, split into passages and kept current. Ask a question and Cortex shows the answer with its citations and match scores — and, just as importantly, which sources it set aside and why. Answers that prove their worth become playbooks: structured, reusable prompts the whole team can run again to the same standard of evidence. Built-in evaluations keep testing that answers stay grounded in their sources, and the usage dashboard shows exactly what the workspace is spending against the budget the team set — per project, per model, and per person — before a limit is ever reached, not after.
This project is the complete Cortex product experience, designed and built end to end: the public site and pricing, the full workspace — conversations, knowledge bases, a passage-level inspector, playbooks and their builder, evaluations, and usage — team settings and billing, and a documentation set that explains the citation system itself. It is built for the moments when an answer has to be forwarded — to legal, to leadership, to a client — and defended line by line.
- Answers with receipts — every claim footnoted to the passage behind it, the source one click away
- Knowledge bases — team documents in collections, split into passages, with freshness signals as sources age
- Passage inspector — see how a document was chunked, why a passage matched, and what was set aside
- Playbooks — proven prompts saved as structured runs anyone on the team can repeat
- Evaluations — continuous checks that answers stay grounded in the sources they cite
- Cost governance — spend tracked live against a team-set budget, per project, model, and person
Engineered like a product, not a demo.
31
screens — the complete product surface, built from approved designs and verified against them screen by screen, at desktop and phone widths alike.