Gnist Credence

AI-assisted reasoning for thinking clearly under uncertainty

Think Clearly About Uncertain Questions

Get the reasoning on paper

Some questions are too tangled to hold in your head — "Should I buy this apartment?", "Will this startup reach its revenue target?", "Will this market resolve yes?" The factors interact, the assumptions hide, and the reasoning gets buried in a long chat thread.

Gnist Credence helps you and an AI assistant build a shared, explicit, probabilistic model of the problem instead of leaving the reasoning trapped in someone's head. It gets the thinking on paper — an object you can inspect, challenge, update, and improve over time.

It is not an oracle. The value is not that the AI knows the answer; it is that the AI helps construct and maintain a structured picture neither of you could reliably hold alone.

Photorealistic interior of a 1960s executive study at dusk. Polished dark walnut desk with brass accents, green-shaded banker's lamp casting warm amber light, a crystal tumbler on a leather blotter, a fountain pen resting on an open notebook. Floating above the desk, a translucent holographic belief network: softly glowing cyan nodes labelled with short yes/no questions, connected by thin directed arcs, each node carrying a small probability gauge. Warm-to-cool color temperature split between the wood-panelled room and the cool data-glow. 35mm lens, f/4, cinematic grading, subtle film grain.

Reason Through an Inquiry

One shared, inspectable model

You start with a plain-language question, not a diagram. From there, Credence builds an Inquiry — a structured investigation that holds the moving parts of your reasoning:

Beliefs — probabilistic statements that feed the question, each tagged as yours or the AI's
Evidence — information that should move those beliefs, traceably linked
Key drivers — the assumptions your conclusion is most sensitive to
Update history — an inspectable trail of every change: what moved, why, and who changed it

The AI is active but never overconfident: it surfaces hidden assumptions and proposes structure, but never silently overwrites your beliefs or collapses the uncertainty into a single confident number.

Photorealistic close-up over a walnut reading table of a single translucent glass panel held upright in a brass stand. On the panel, an Inquiry takes shape: a central question card reading 'Should I buy this apartment?', surrounded by smaller belief cards each showing a probability between 0 and 1, some tagged with a small human icon and others with a subtle AI spark icon. Thin cyan arcs link the beliefs to the question; one belief is highlighted as a 'key driver'. A brass-rimmed magnifier rests beside the panel. Warm desk light, cool panel glow, shallow depth of field. 50mm lens, f/2.8, cinematic grading.

Early — and Honest About It

Concept to MVP and beyond

Gnist Credence is at the concept / pre-MVP stage. A working scaffold exists; the reasoning maths are deliberately minimal while we get the core experience right.

Credence is AI-native first: the primary interface is an MCP server your AI client connects to, with a REST counterpart for testing. You can start anonymously with a lightweight token and later attach your work to an account without losing it.

Prediction-market forecasters are an early proving ground — they already think in probabilities and can measure their calibration over time. Richer evidence handling, more structure, and portability across AI clients are on the roadmap.

Photorealistic three-quarter view of a 1960s executive desk where a warm wood-panelled world meets a cool neo-futuristic one. A monitor on a brass arm shows a dark-mode chat client connected, via a glowing cyan link labelled 'MCP', to a small probabilistic model rendered as nodes and arcs. On the desk: a mechanical keyboard with walnut wrist rest, a brass desk lamp, a notebook with hand-drawn probability sketches. Through a half-open wooden door behind the desk, a bright white corridor with cyan LED baseboard is barely visible. Afternoon light, warm grading with cool accents. 40mm lens, f/3.5, gentle background bokeh.

Try Gnist Credence

Bring your own AI client

Connect your AI client and start turning a fuzzy question into an explicit, shared model you can actually reason with.

This is an early prototype — expect rough edges, and tell us what breaks.