Iteration 2 re-shells the whole Talk / Book / Connect set into a left sidebar (the ChatGPT / Claude / Gemini convention users now expect) — sidebar = your brains, conversations & nav; main pane = the chat/content. Grounded in staging's existing thin left icon rail and evolved into a wide, labelled sidebar. Same warm system throughout. Plus the missing bridge: a matched-results → expert screen between "ask a question" and the answer.
The staging-faithful "One question, many minds" ask hero now lives in the main pane; the evolved sidebar carries search/discover, your recent brain conversations, and Book/Connect/Dashboard — the ChatGPT-style shell.
Open →The screen R2 was missing between "ask" and "answer." Keeps "One question, many minds": your question echoes at the top, then a few best-fit brains each with a "why matched" line and Talk / Book / Connect affordances. Doubles as browse/discover.
Open →Talk is the hero (grounded chat, citations, honest→book handoff). Proper Book and Connect sections sit alongside — Connect surfaces "add this brain to your apps."
Open →A stranger arriving from a shared link sees the same sidebar shell, but it adapts — Explore/discover + a "Create your own brain" CTA + Sign in, not an empty conversations list. Main pane still foregrounds talk / book / connect / sign up.
Open →"Add to Claude / Add to ChatGPT" → a permission modal (the trust moment) → connected. Honest that a fully-silent install isn't possible; consent stays.
Open →When someone books through a connected brain inside Claude: availability surfaces in the host, they pick a slot, then a handoff brings them to a Marnji confirm-and-pay page (Stripe).
Open →A dedicated surface to get an expert's brain out there — copy link, social share, a QR, and an embed snippet for their own site.
Open →Solid, ongoing materials workflow (add/remove sources + re-index), what the brain confidently knows vs defers on, and the three reach doors (Public / Book / Connect).
Open →Left sidebar shell, everywhere (Khoa's call). R2's top-nav is replaced by a left sidebar on every surface — the convention every AI app now uses (ChatGPT / Claude / Gemini), and a natural fit for a chat-centric product. I grounded it in staging's existing thin left icon rail (toggle · new-chat · discover · help) and evolved it into a wide, labelled sidebar: New question · Search & discover · Recent conversations (brains you've talked to) · Your brains · Book / Connect / Dashboard · profile at the foot.
Context-aware, not two shells. The public brain page (a logged-out stranger from a shared link) keeps the same shell visually, but the content adapts — Explore + "Create your own brain" + Sign in instead of a personal conversations list. One consistent shell, content that knows who's looking.
The missing bridge is in. A new matched-results → expert screen sits between "ask a question" and the answers page: your question → a few best-fit brains (each Talk / Book / Connect) → pick one → talk. "One question, many minds" is preserved and finally has its results moment.
Everything else from R2 is kept — just re-shelled. Talk/Book/Connect on the answers page, the add-to-app click-once + one-tap consent modal, the booking-via-host handoff (initiate in the host → complete & pay on Marnji), the share surface, and the owner dashboard/data-management are all intact inside the new shell. Same warm system (ink #181614, Fraunces + Figtree, cream/terracotta, arch avatars, pill buttons); no end-user MCP jargon.
No end-user jargon. Users never see "MCP" — they see "add to Claude / your apps." The mechanism is invisible; the outcome leads. No Spotify/royalty language anywhere.
Two honest constraints made visible (per your realism note): (1) connecting can't be fully silent — the one-tap consent modal is the trust moment, and tapping "Add" deep-links into the host's own connector flow; (2) booking through an external app initiates in the host but completes and pays on Marnji (Stripe), so money + details stay secure.
Marnji · Iteration 2 — left shell nav + matched-brains bridge · design-only, for build by Ben · Sana → Maya · 2026-07-23 · iterates on the R2 set, grounded in staging.marngi.com's existing left rail.