About thinksn

What is thinksn?

thinksn helps you build a position on a hard question — and shows you where that position is strong and where it is exposed. It is not a search engine, not a chatbot, and not a mind map. It is a place to think through a question slowly, with the structure of your reasoning visible while you work.

Instead of returning an answer, thinksn grows a network of claims with structure around your question. Each claim is generated under a specific analytical frame and tested through specific moves — including moves that try to break it. What emerges is not a summary of a topic. It is a working argument with its joints visible: the grounds it rests on, the assumptions doing the heavy lifting, and the places where the reasoning is single-threaded rather than convergent.

The network is not meant to be complete or authoritative. It is a structure you build under deliberate constraint, one expansion at a time, so that when you walk away with a conclusion you also walk away knowing what it costs you to hold it.

A sample network is the fastest way to see what this feels like in practice. Click, expand, replace, and follow the reasoning where it leads.

How it works

Every question is opened along a small number of frames — distinct inferential dimensions through which a claim can be made. A frame is not a topic or a tag; it is the kind of argument being attempted. Causal frames ask what produces what. Comparative frames ask how cases differ. Evaluative frames ask whether something meets a criterion. Predictive, historical, definitional, prescriptive, and several others each carry their own logic.

Inside any frame, you can apply lenses — directed moves on a claim. Some lenses add support: an example grounds a claim in a concrete instance, a mechanism traces the process connecting evidence to conclusion. Others are adversarial by design. A counter move argues for a rival mechanism entirely — a different explanation that would displace the one you're holding, not just a hedge on it. A blindspot move grants your evidence but challenges whether it actually licenses your conclusion. A context move asks whether the conditions your reasoning depends on are actually the ones in force here.

These lenses are not equal viewpoints. Counter, blindspot, and context are the system pressing on the argument — asking a real question and returning a real, adjudicated answer: did the claim survive as stated, survive narrowed, get absorbed into something better, or get replaced outright. Their findings are what tell you whether a conclusion is load-bearing or merely unchallenged.

At every step, only a limited number of concepts can appear. Expansion is intentional; nothing accumulates without a choice. This is what keeps the network legible, and it is also what makes the resulting argument yours rather than a transcript of the model talking to itself.

What thinksn adds

Most hard questions live inside a network of related concepts, competing frames, and hidden assumptions. We navigate these networks naturally, but the navigation is usually implicit — shaped by available vocabulary, by what we read most recently, and by the conclusions we are already inclined to reach. thinksn makes that navigation explicit and adversarial. The point is not just to see what you are already thinking, but to find out what your thinking cannot survive.

What this looks like in practice is a set of visible attack surfaces. Frames let you see which inferential dimensions you have actually opened and which you have left untouched. Lenses let you see which claims have been grounded, which have been stress-tested, and which are still sitting on a single thread. The same network read two ways: as a map of what you have established, and as a map of where the argument is exposed.

The report reads the network in exactly this register. It does not summarize. It identifies what the argument rests on, where its strongest chain of reasoning runs, where that chain is single-threaded, and where it has survived a direct challenge.

Who it's for

thinksn is for people who want to take their own thinking seriously — not only explore a question, but understand what their reasoning rests on. Researchers working toward a claim rather than around a topic. Writers who want to know where their argument is single-threaded. Anyone who would rather find their own weak spots than have someone else find them first.

It is most useful at the stages where a position is still taking shape, when the difficulty is knowing which questions are worth asking and which assumptions are doing the heavy lifting. It will not resolve ambiguity, but it can make ambiguity navigable: clearer in its structure, more honest about its tensions, and easier to work with deliberately.

A few things are worth knowing upfront. The methodology structures questions differently than search or conversation, and the outputs will sometimes feel unfamiliar — that is partly intentional. The AI layer works best as a thinking partner with known limitations rather than an authority: it can generate plausible-sounding connections that do not hold up, and it currently operates without internet access, so rapidly evolving topics may reflect an earlier state of the world.

Where this is going

What you're using today for free lets you build out a network by hand — asking the next question, following up on what you find. It's the same underlying machinery Pro builds on, not a stripped-down preview: every claim you generate is real, structured reasoning, not a demo. What free doesn't include yet is the adversarial side — attacking a claim, comparing rival explanations, constraining or steering the reasoning, and grounding it in outside sources. That's what Pro adds.

Pro is coming, aimed at the moment when the question matters and the answer will be challenged — by a committee, a client, a reviewer, or your own second reading a week later. It adds an automated investigation mode that runs the full cycle — building out an answer, attacking it, comparing it against rival explanations — across many rounds without you clicking through each step; grounding in your own uploaded documents; and reports that read the resulting graph back as a stress-tested case, not a summary. We're aiming for a beta by late 2026 — a real target, not a promise; the usual things could push it back.

You don't have to wait to see what that full version produces, though. A growing set of finished investigations — run all the way through, attacks and rival comparisons and all — is already there to browse, free or not. It's the fastest way to see what the automated version actually looks like before deciding whether it's worth paying for.

Until then, the free version is what's here: real reasoning, built one deliberate step at a time, with you deciding what to push on next.