the frame challenged.
the structure surfaced.
some thinking needs somewhere to go.
jeves is a thinking environment, not a chatbot. Consumer AI flatters and aligns to your framing. Jeves does not. It is an AI thinking partner built for hard decisions, half-formed ideas, and recurring situations whose structure you cannot see from inside. The kind of operation changes with what you need, not just its intensity.
It does the right thing for where you are.
When you are working through a decision, it challenges the frame. When you are generating ideas, it follows and extends, it does not evaluate. When you are processing something hard, it stays with what is real for you. Same system. Different operation depending on what you need.
Not one mode applied to everything.
It surfaces what is underneath, not what is on the surface.
The most useful thing to surface in a conversation is usually not a reaction to what you said, it is the structure underneath it. A hidden assumption. A feedback loop you are inside of. A frame that is producing the difficulty. Jeves looks for those before responding to the surface. When your reasoning contains a distortion, it names it in three sentences, with an analogy fitted to your situation, and a reframe.
Latent structure, not surface response.
It does not simulate certainty it does not have.
When the evidence is one-sided, Jeves preserves multiple readings rather than collapsing to one. When a claim about your own behavior or motivation is not grounded in what you actually said, it does not make it. Uncertainty is not a weakness in a thinking system, it is structural integrity.
Epistemic honesty. Not hedging.
- you are working through a hard decision and want the frame examined
- you are generating ideas and want them extended, not evaluated
- you are inside a recurring situation and can't see the structure of it
- you want something named that you have been circling around
- you need fast factual answers
- you want to be reassured rather than oriented
- you want to be told what to do
- you want help with a routine task
most of what jeves does per message is invisible. here is the part you never see.
- ◇every message is checked against a library of forty reasoning patterns — logical fallacies and cognitive distortions — each scored for how load-bearing it is to your argument. only the structural ones get named.
- ◇claims jeves makes about you must be grounded in your own words. before a response ships, each one is checked against what you actually said — sentences that fail are deleted.
- ◇jeves holds up to four parallel readings of what you mean. a reading collapses only when evidence rules it out, not when it merely falls behind.
- ◇each turn, jeves predicts where your reasoning is headed — with a condition that would falsify it — then grades itself against your next message. the patterns window keeps score.
- ◇challenge is mode-aware. in exploring mode, fallacy flagging is suppressed on purpose: adversarial pressure on a generative state collapses it.
- ◇the pentagon is not decoration. the state you report blends weighted directives straight into the model — where you put the shape changes how hard jeves pushes.
- ◇what jeves remembers about you is a set of weighted hypotheses, not facts. each decays over weeks unless re-evidenced, and all of it is readable and erasable in the memory window.
- ◇jeves does not align with your mood. if the frame is wrong, it says so. if the question is confused, it names the confusion before answering it.
- ◇when you describe a conflict, jeves treats your account as one side of it. claims about the other person's motives stay inferences — never facts.
- ◇attached documents are consumed by a single turn and then deleted from the server. nothing you upload is kept.