The interface that adapts itself
for every visitor.

No. SentientUI runs the learning API, optimizer, visitor profiles, audience grouping, and graph storage on our infrastructure (api.sentient-ui.com). You install @sentientui/react, add API keys from the dashboard, and point env vars at our hosted endpoints. There is no customer-facing self-host path today — you focus on your React app; we operate the learning loop.

Feature flags and classic A/B tools expect you to configure experiments, read dashboards, and ship winners manually. SentientUI is a dynamic software interface layer: your sign-up, pricing, and onboarding pages learn who each visitor is and adapt automatically — no experiment manager, no analyst making the call. The AI optimizer, visitor profiles, and audience groups run on our hosted API; you just declare your variants in JSX.

SentientUI runs an AI optimizer per component, per audience segment. It maintains a probability distribution over each variant's conversion rate and samples from it to decide which to show. Uncertain variants (less data) are naturally explored more; confident winners are served more often. The algorithm self-tunes without a fixed exploration schedule — you never need to declare a winner manually, though you can promote one at any time.

A portrait is a real-time behavioural profile built for every session. It tracks six dimensions — urgency (how quickly the visitor moves toward conversion), attention span (dwell time × scroll depth), decision stage (how deep into the funnel they are), price consciousness (time spent on pricing vs total dwell), risk aversion (reliance on trust signals), and social influence (whether social proof drives them to act). Each dimension is a 0–1 score that updates as events are processed, typically within minutes. A reliability score (also 0–1, based on event count) tells you how much weight to give the portrait — new sessions start at 0.5 defaults and converge toward true behaviour as data arrives.

Once a session has a complete visitor profile, a nightly job assigns it to one of four audience groups: buyers (high urgency, ready to convert), researchers (high attention, thorough decision-making), deal-seekers (highly price-conscious, heavy comparison), and browsers (exploratory, low urgency). From that point, the optimizer serves segment-specific results for that audience — so buyers and researchers can have different winning variants from the same experiment. Projects with fewer than four complete profiles use device + traffic source only.

When you enable graph mode, the SDK scans the page and discovers structural and semantic relationships between your components — for example, hero → CTA → pricing. These connections are stored server-side. When a goal fires on one component, the system walks the graph up to two hops and credits related variants with a weighted share of that score. A hero headline can earn partial credit for a pricing conversion it influenced downstream. This makes the optimizer learn faster on low-traffic components by borrowing signal from high-traffic neighbours. The dashboard shows each variant's direct vs shared score split.

Yes. Wrap your root layout with <AdaptiveRoot> and list your component IDs. Assignments are resolved server-side before the HTML is sent to the browser — variants appear in the initial markup with no layout shift, no loading flash, and full crawlability for search engines.

A session is one unique visitor in a calendar month, identified by a first-party cookie (_snt_uid) containing a random UUID. If the same visitor returns multiple times in the same month, it still counts as one session. The counter resets automatically on the first of each calendar month.

Yes. The _snt_uid cookie contains only a random UUID — no personal data, no fingerprinting, no cross-site tracking. Visitor portraits are computed from anonymous behavioural signals (dwell time, scroll depth, event patterns), never from personal identifiers. If you need explicit consent before setting cookies, pass consent={false} to AdaptiveRoot or AdaptiveProvider and flip it to true once the visitor accepts. The SDK does nothing until consent is granted.

Session IDs (random UUIDs), device class, traffic source, UTM parameters, which variant was shown, goal events, dwell time per component, and scroll depth. From these signals the API derives visitor profile scores — all computed server-side from anonymous behavioural patterns, with no personal identifiers stored. No IP addresses are retained. Raw events are purged after 90 days; aggregated optimizer weights and visitor profile data are retained until you delete the project.

Yes. Use <AdaptiveProvider> from @sentientui/react in any React app — Vite, Remix, Create React App, Next.js Pages Router. Assignments are made client-side after mount in those environments (no SSR preloading), but the AI optimizer, visitor profiles, audience grouping, and graph score sharing all work normally.

New session creation returns a 402. Visitors already on your site can finish their session normally — existing sessions keep sending events, the optimizer keeps learning, and visitor profile computation continues. The counter resets on the first of the next calendar month and new sessions are accepted again automatically.

Yes. The dashboard shows accuracy ranges and significance markers for each variant. Once you are satisfied with the data, click Promote — all future traffic goes to that variant regardless of segment. You can unpromote at any time to resume optimization.

AI Insights are plain-English observations about what changed in your project this week — generated automatically from your conversion data. Every project page opens with 3–5 narrator bullets covering variant conversion rate shifts, audience velocity, session volume, and any guardrail events. Growth plans also get 1–2 advisor recommendations on what to do next. Insights are generated using Claude Haiku and cached for 6 hours. Hit Refresh to generate fresh insights at any time. Narrator and project chat are available on Starter and Growth plans (5 chat queries/day on Starter, 20/day on Growth); advisor recommendations are Growth-only.