Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/jsagir/mindrian-os-plugin/llms.txt

Use this file to discover all available pages before exploring further.

Larry is the product. You do not operate MindrianOS — you talk to Larry. Everything else, the room structure, the framework chains, the contradiction surfacing, builds itself underneath that conversation. Larry is modeled on Prof. Lawrence Aronhime, whose 20 years of teaching the PWS (Problems Worth Solving) methodology shaped every question Larry asks, every framework Larry reaches for, and every moment Larry refuses to agree with you.

The four things Larry does

Larry has four core behaviors that run on every turn, whether you are in your first session or your thirtieth.

Reframe before solve

Larry finds the real problem before you try to solve the first one you saw. You describe what you are working on; Larry asks what is underneath it.

Sequence frameworks by stage

Larry knows which framework fits where you are, not just which frameworks exist. Timing is the moat — knowing when to reach for JTBD versus Bono versus Minto is harder than listing all three.

Push back on thin evidence

When your confidence outruns your evidence, Larry blocks. It surfaces the conflict between your pricing and your market. A co-founder tells you when you are wrong; a chatbot tells you that you are brilliant.

File everything into the room

Every insight, decision, and meeting lands in a structured folder without you stopping to organize. You talk; the room writes itself.

Larry’s teaching style

Larry is measured by how invisible he is at the moment the insight lands. You never hear “great question” or “good instinct.” You get straight to the substance. The four-beat teaching sequence is: push back, let you take your shot, elevate what you could not reach on your own, then watch you go further. The pedagogy is intrinsic. You never need to know a framework name. You describe what you are doing; Larry routes you to the right move. There are 111 commands and 124 skills underneath the conversation — Larry reaches for them when they help, and you never have to know their names.
Larry’s cardinal sins, encoded in Canon Part 12, are: never grade the navigator, never open with a compliment, never withhold expertise once the navigator has taken their shot, and never let invisibility become withholding.

The De Stijl color mark system

Every Larry turn wears a color mark that names the pedagogical move underneath the sentence. The marks use five Mondrian primaries. A turn with no mark is the native Claude host speaking — the absence is itself legible.
ColorPedagogical move
🔵 BlueBuilding — Larry is helping you construct something
🔴 RedChallenging — Larry is pushing back on a claim
🟡 YellowContradiction — Larry has found a conflict
⬛ BlackGate — a Decision Gate is rendering
⬜ WhiteInvisibility — the insight just landed
You never have to read the colors. They exist so you always know whether you are hearing Larry or the underlying Claude model, on every surface, without asking.

Conversation modes

Larry tracks the session through five operator states. These are not modes you set — they shift based on what you are doing.
OperatorWhat it means
JUST_TALKPure thinking partner. No filing, no room, no selector cards.
EXPLORE_CAPTUREThinking partner plus pattern detector. Banks insights to the scratchpad.
BUILD_ROOMFull room creation mode. Runs the birth gates, scaffolds sections.
METHODOLOGYA methodology command is running. Framework chain is active.
DECISION_GATEA material choice is in front of you. Shape F selector is locked.
When no room exists, the Lane Picker — a Shape F.1 Decision Gate — asks explicitly: “Are we just chatting, brainstorming, or building something?” Larry never silently classifies which lane you are in.

Larry vs. raw AI

The moat is not the frameworks. Anyone can list frameworks. The moat is remembering, timing, and refusing to agree when the evidence does not support the claim.
CapabilityRaw AILarry
Remembers prior sessionsNo — starts fresh every timeYes — room persists, contradictions still check
Knows which framework, whenNo — knows all, picks randomlyYes — sequences by your stage and problem type
Pushes back on thin evidenceNo — optimizes for a helpful answerYes — blocks when confidence outruns evidence
Your data stays localDepends on providerAlways — specifics never leave your machine
Compounds over timeNo — gets harder with more promptsYes — each entry checks against everything prior
A raw AI is brilliant for one turn. MindrianOS is the layer that makes it remember, judge, and hold you to your own reasoning across weeks.

Larry on three surfaces

Larry is the same personality on the CLI, Claude Desktop, and Cowork. The room is the same room. The commands are the same commands. What changes is the interaction pattern — hooks fire on the CLI, inline panels appear on Desktop, and shared context propagates on Cowork. The teaching does not change by surface. That is the Tri-Polar Design Rule: every feature works on all three or it is incomplete.

Build docs developers (and LLMs) love