The Socratic Guide
Most AI does your thinking for you. Neural Garden’s guide grows a map of questions worth opening — and leaves the thinking with you.
Expand any thought and three suggested nodes fan out — one always an angle you hadn’t considered.
There are two ways to put AI next to a thinking person. One replaces the thinking. The other — the one great teachers have used since Socrates — points at paths and lets you walk them.
Hands you a finished map. Impressive, hollow — the skill it exercises is its own.
Opens three doors and asks which pulls you. Understanding you construct is understanding you keep.
Five old ideas about how people learn — each one a design decision in the garden.
Socrates
Understanding is drawn out by questions, not handed over.
In the garden — Three openings at a time, ending with a question. You do the concluding.
Lev Vygotsky
Learning happens one reachable step beyond what you can do alone.
In the garden — Each tier grows from a thought you wrote. The frontier stays one step away.
Piaget · Slamecka & Graf
Knowledge you construct yourself is the knowledge you keep.
In the garden — Suggestions are sketches until you keep, rename, or write into them.
George Loewenstein
Curiosity ignites at the gap between known and knowable.
In the garden — Titles are hooks that make you want to open the branch.
Stuart Kauffman
Growth is exploring what sits one step away — new doors open as you walk.
In the garden — Every tier slips in one branch you didn't know you didn't know.
A suggestion is only as good as its context. Each tier is drawn from everything the garden knows — including what you said no to.
Every suggestion knows where it stands: the parent it must fit, the siblings already explored (and what they collectively miss), the children to go deeper than — never a thought in isolation.
Write a goal for the garden and every suggestion steers toward it. The goal is yours alone — the AI reads it everywhere but never rewrites it.
Socratic questions by default — but a planning branch gets concrete next actions, a stuck question gets a compact answer, an untested assumption gets a counterpoint.
Rarely — and only when it matters — a differently-tinted note appears about your thinking itself: a branch running deep with no synthesis, siblings circling one angle, drift from your goal.
Turn on Pre-think and the frontier's next paths are prepared while you're not looking — two levels deep. Select a thought and its three doors are already open; step through one, and the doors behind it are open too.
Every dismissal is kept on that thought — across sessions. The guide never re-offers a rejected angle, and each ✕ quietly teaches it where you don't want to go.
No mode to switch on — exploration is simply how the assistant responds when you’re exploring.
Press + on a thought — right for where it leads, left for what it requires — or tell Cmd+K to "expand this" or "explore prerequisites". Commands get done; exploration switches the AI into guide mode. No toggle.
Placeholders shimmer instantly, then fill in as dashed suggested nodes — real, movable, already connected. Direction is the meaning: downstream grows the idea, upstream adds what it rests on.
✓ makes it yours. ✕ prunes it — and a fresh path grows in its place. Editing a suggestion adopts it instantly.
One tier at a time, deeper levels unopened. Expand what pulls you — a skill tree, a learning path, a life map. Dense branches become their own gardens.
A solid connection. Now part of your garden — and your search, clusters, and intelligence.
It disappears — remembered as a no, never re-offered — and a fresh path grows in its place, so the branch keeps offering.
Renaming or writing into a suggestion adopts it instantly. Making it yours IS accepting it.
No mystery box. This is what your garden gathers, what the AI is actually asked, and what comes back — so you know exactly what to expect from every expand.
Everything on the left is gathered locally from your garden. One request goes to the AI provider you configured, and both of your decisions — keep and dismiss — flow back into the context for next time.
Every neighbor is labeled by relationship: the parent is a frame to fit, siblings show what's already covered (and what they collectively miss), children mark where to go deeper or diverge. The branch path root → here sets the altitude — deeper means more specific, never back to generic.
One next step along this exact path. One alternative angle heading a different direction than the existing children. One adjacent surprise you didn't know to ask for. Upstream (the left +) flips the roles: a prerequisite, a broader frame, a hidden assumption.
Socratic questions are the default — but the guide is told to switch when the context calls for it: a concrete next action on a planning branch, a compact answer when your thought poses a direct question, a counterpoint when an assumption goes unexamined. The node type follows the form.
Titles are 3–5 words, like map labels. Bodies are one line, under ~15 words. Openings, never essays — the thinking stays yours.
Everything already on the branch, nearby on the canvas, previously suggested, or dismissed by you is listed as off-limits. Your no's are permanent context — a rejected angle stays rejected.
After drafting, the guide assesses the thinking itself against your goal — depth without synthesis, siblings circling one angle, drift from what you said you're after. Only when a signal is clear may it add ONE coach note. Most tiers: none.
Guardrails, not vibes — each one is enforced in the product.
Suggestions stay out of search, clustering, and garden intelligence until you accept them.
Every suggestion is visibly tentative — dashed, frozen, one click from gone.
A good question starts three thoughts; a finished answer ends one.
The guide clears the path. You walk it — that's why it sticks.
Write down the thing you’re curious about, press +, and see which of the three paths pulls you.
Try Neural Garden