The AI study platform for medicine and science.
General chatbots answer from memory. Stella answers from the textbooks your course is taught from, verifies current guidelines on the live web, remembers what you keep getting wrong — and tests you on it until it stops being wrong. Built for MBBS & USMLE, BSc and MSc. Free to start.
General AI answers. Stella manages learning.
A general-purpose assistant — ChatGPT, Claude, Gemini — is a trained-once oracle. It answers from what it learned on the day it was built, it offers no source you can check, and it forgets you the moment you close the tab. For a lookup, that is fine. For learning, it is the wrong instrument: it doesn't know your syllabus, the format of your exam, or where your understanding is weak.
The difference is structural, not cosmetic. Before Stella writes a word, the platform retrieves the relevant passages from the textbooks your course is actually taught from — and shows them to you. It checks the live web when a guideline has moved since training, and cites what it found. It assembles a briefing from three tiers of persistent memory — who you are, what's shaky, what's settled. And it teaches the way your exam tests: mechanisms first, traps named, questions asked mid-flow rather than bolted on at the end.
That is a platform built for one job — learning at examination standard — not a chat tool with a study skin. The models underneath are the best available; the retrieval, verification, memory, routing and test engines around them are Stella.
Any chatbot answers. Stella makes sure you keep it.
Six systems that separate learning from lookup.
None of this is a clever prompt. Every answer passes through retrieval, verification, memory and routing before a word is written — so what reaches you is grounded in a real source, current, and addressed to you specifically.
Answers from the books you're taught from
Stella is wired into curriculum textbooks, partitioned per programme. She retrieves the relevant passages first, answers out of them, and shows you the source — so you can check her against the page.
Current against this year's guidelines
A model's knowledge stops on the day it was built. When an answer turns on a threshold, a first-line agent or a protocol that has moved, Stella reads the live web before replying — and cites what she used.
A model of you that compounds
What you're working on, what's unsteady, what's settled — kept in three tiers, decayed and re-ranked on a schedule, and folded into every reply. You never re-introduce yourself.
The best available model, with failover
Answers are routed across providers by task and quality tier. If a provider stumbles mid-answer, the turn moves to another lane — and a retried answer is never billed twice.
She executes, not just explains
Hand her a topic and she plans the mission, sweeps the web and the library, then writes complete study materials into your workspace — research briefs, exam packs, decks, schedules — with a receipt of what it cost.
Built around how examiners think
Stems in board shape, distractors named and autopsied, viva questions anticipated, and every miss returned on a spaced schedule until it stops being a miss.
The session, end to end.
A conversation with Stella is a managed process — briefing, retrieval, teaching, testing, filing — not an open-ended chat. Give her a topic and a sitting, and she runs it in pieces, out loud, with you doing the work.
Context is assembled before you type
Your programme, your profile, the last session, the concepts you're shaky on — gathered before your question reaches the model. You just type the question.
Evidence is gathered first
The relevant passages are retrieved from your textbook library, and the live web is checked where guidance may have moved. The answer is built on sources you can see.
Teaching is shaped to your gaps
One register, one effort level, one depth — chosen for the moment, and tuned by subject: cardio gets cardiology-shaped guidance, pharm gets pharm. Say slower or I don't buy that and she takes it apart further.
Testing happens mid-stream
Questions arrive while the mechanism is still warm — not as a quiz bolted on at the end. A wrong answer is autopsied: why the distractor was written, and what it caught you thinking.
Everything returns on a schedule
Misses are filed to your knowledge state, scored, and brought back on a spaced schedule — and into tomorrow's plan. You never have to decide what to revise; the misses put themselves back on the list.
One topic. One sitting. Finished — and filed, so tomorrow starts where you left off.
Stella remembers what you're learning, not just what you asked.
Three layers, each running on its own clock: what's live right now, what's unsteady this week, what's settled for good. Nothing has to be said twice, and nothing is judged once and forgotten.
Session context
The case in front of you. Labs pasted four minutes ago, the mechanism half-traced, the distractors still standing. It rewrites itself as you think out loud.
The map of what's unsteady
Every concept you touch carries a state — solid, unsteady, or weak — scored, decayed over time, and re-ranked by subject. This map is what decides what tomorrow looks like.
The permanent record
The settled reasoning, kept for good and searchable. The next conversation opens already knowing what the last one worked out — no re-introductions, no repeated ground.
Context that never expires.
Between turns, the compaction engine reads the conversation back, keeps the reasoning, and writes a compact ledger entry in its place. The window never fills, so the thread never gets cut off — sessions run for weeks, not forty messages.
Nothing skimmed
The full exchange is re-read between turns — every value you pasted, every branch you eliminated — not a rolling window that quietly drops the start.
Compressed, not lost
What mattered is distilled into a searchable ledger entry — the derivation survives, the token weight doesn't.
No forced stopping point
Elsewhere the ceiling arrives twelve exchanges deep, on the turn the reasoning finally starts to close. Here the context renews itself between turns, so there is no built-in end to a working thread.
One core. Three specialised engines.
Interpret the report, not just the values
Paste a raw report and Stella marks what's off, names the pattern, ranks the differentials that could produce it — and tells you which two values the examiner will make you compare.
Questions built from your own thread
Type /test and the block is generated from the topic you just finished — not pulled from a generic bank. Every miss lands on your knowledge state and returns in a later block until it stops being a miss.
Cards that test two ways
Highlight a line and it becomes an SM-2 card, run blank-recall or disguised as options. Fail one and what comes back is the mechanism note you were missing — not just the answer you should have picked.
Autonomous missions. Complete study materials.
One toggle, one topic. Stella plans the mission, probes your textbook library, sweeps the live web — and writes complete, polished study materials into your personal workspace. You watch the whole run, and there is a Stop control at every moment.
A current, cited brief
Any topic you're studying, swept against the latest guidelines and trials, sources numbered and cited, shaped for exam relevance. Recency first, always.
Notes you'll actually reread
A self-contained study sheet with elegant typography — comfortable on phone, tablet and desktop — or downloaded and kept. Dark mode included.
Question-first revision material
Built the way examiners actually test a topic — stems, distractors and the reasoning that separates them, drawn from your programme's exam shape.
A schedule written into your calendar
Tell her the exam date and the gaps; she writes day-by-day checklists straight into your Daily Plan — sequential, realistic, and aware of your weak map.
Live timed MCQs in the chat
Miss a question and she drills the gap, then teaches it back. A second engine keeping the first honest.
Full transparency, always
Every file lands in your workspace — your own notes too, for her to read. Every mission ends with a receipt: exactly what it used, shown as a share of your plan. No mystery billing.
One question. Four response registers.
Some moments want the whole mechanism, some want six lines before rounds, some want to be examined. Switch register mid-thread — the memory stays exactly where it was. Teach, Explore and Flash come with every programme; Vignette is Medicine's, built around the case.
Mechanism
digoxin binds Na⁺/K⁺-ATPase → K⁺ competes for the same site → hypokalaemia removes the competition → more pump inhibition → Ca²⁺ overload → arrhythmia
Digoxin and potassium fight over the same molecular doorway on the pump's extracellular face. Fall the potassium, and digoxin binds freely — the Na⁺/K⁺-ATPase slows, intracellular Na⁺ climbs, the sodium–calcium exchanger runs in reverse, and the myocyte's calcium rises until delayed afterdepolarisations fire.
The trap
The examiner will offer hyperkalaemia as the aggravator. It is the opposite: potassium is the natural antidote — which is why the first-line therapy is to give it.
Because potassium and digoxin bind the same site on the sodium–potassium pump, and potassium wins when it's present. When serum potassium falls — say, from a loop diuretic — digoxin binds more readily and the pump slows further.
The consequences cascade: the cell can't push sodium out, so sodium–calcium exchange reverses and calcium accumulates, the myocyte becomes excitable out of turn, and you get the classic rhythm — ventricular ectopy, bidirectional VT, and every toxicity feature from nausea to xanthopsia.
K⁺ and digoxin share the pump's binding site.
Hypokalaemia → digoxin binds freely → pump slows → Na⁺ rises.
NCX reverses → Ca²⁺ overload → afterdepolarisations.
First-line: give K⁺. Hyperkalaemia never worsens it.
A 68-year-old on digoxin presents with nausea, yellow-tinted vision and diarrhoea. K⁺ 2.9 mEq/L. ECG shows bidirectional ventricular tachycardia. Which intervention is correct?
A — IV calcium gluconate
B — IV potassium supplementation
C — Double the digoxin dose
D — Cardiovert immediately
Answer · B
Hypokalaemia is the trigger and potassium the first-line antidote — it outcompetes digoxin at the pump and restores the gradient. IV calcium is dangerous in digoxin toxicity, and cardioversion is reserved for refractory haemodynamic instability.
The syllabus changes. The platform doesn't.
A programme decides the shape of a question, the sources it's answered from, how the answer is spoken, and which study tools come with it. It never decides what Stella remembers about you. You choose once when you sign up, and it stays.
MBBS, and Step 1 & 2 CK
Stems arrive in NBME shape, answered mechanism-first because the exam rewards the derivation and punishes the memorised fact. Anatomy through forensic medicine, shaped the way an Indian university examines — the written paper, the specimen on the table, then the questions answered aloud while someone watches you think. Sit both — the memory doesn't move.
The science degree, across six disciplines
Chemistry, biology, physics, mathematics, computing and psychology — read across the semester the way it is actually taught, with internals and the end-semester paper rather than one distant finals week. The notebook, the practical viva and the error bar are treated as the coursework they are, not an afterthought. Exam Focus when the paper is close; Lab & Practical when the write-up is.
The year the reading gets harder
Research methods, statistics you have to defend, and papers read for what is wrong with them rather than what they claim. Experimental design, data analysis, bioinformatics and research ethics, plus dissertation-shaped work — design, analysis, write-up. Switch between Project & Research and Critique & Appraisal as the week demands. One line Stella will not cross: she will argue with your thesis. She will not write it.
The day is built from your knowledge state.
Ask for a day — or a whole exam run, days at a time — and Stella reads your weak map and your recent ledger entries before writing it: what's overdue, what turned shaky, what has finally earned a test. Each day arrives as a card you approve, then lives on the calendar as a checklist. Whatever you don't tick off is handed to tomorrow as unfinished business.
Every plan is the whole product.
Free is the real Stella on a smaller allowance — not a crippled demo. No feature is locked behind a tier, and the same reasoning model runs in every plan. Three things change between plans: how much you can use it, how deep the memory runs, and how much headroom the agent gets. Usage is a weekly allowance plus a flex reserve — never a five-hourly cliff.
Every programme costs the same. A few tools belong to one curriculum rather than to a price tier — Lab mode and the Vignette register are part of Medicine.
- Core long-term memory — Stella remembers your essentials
- Stella Agent included — unlimited runs
- Same reasoning model as every plan
- Deeper memory — Stella remembers much more about you
- A lot more room for Stella Agent — research & study decks
- Faster responses when Stella is busy
- Built for heavy, everyday users
- Deepest memory — Stella remembers the most
- Maximum room for Stella Agent, heavy research days
- Top response priority when Stella is busy
Change tier or cancel from the dashboard, any day of the month
Questions, answered straight.
A curriculum-grade AI study platform for medicine and science. Stella answers from the textbooks your course is taught from, verifies current guidance on the live web, keeps a persistent memory of your progress across sessions, and tests you on it. She is not a general-purpose chatbot with a study skin.
Four structural differences, not cosmetic ones. Grounding — answers are retrieved from your curriculum's textbooks and shown to you, not generated from memory. Recency — when guidance has moved since training, she reads the live web and cites it. Memory — three tiers of persistent, scored knowledge about you, decayed and re-ranked on schedule. Pedagogy — stems, traps and distractor autopsies written to examination standard. A general assistant does none of these four things, because it wasn't built to.
Stella does not train on your conversations. Any deck you build exports to Anki whenever you want, and a chat downloads as Markdown from the message itself.
Yes. You get the same reasoning model and the same platform, with a smaller allowance — enough to take one topic properly from start to finish, which is the only honest way to judge it. Nothing asks you for a card.
The compaction engine reads it back between turns, keeps the reasoning, and writes a compact ledger entry in its place. That is why a working thread with Stella doesn't have a built-in end — sessions run for weeks, not forty messages.
Three programmes, each with its own subject map, sources and voice. Medicine — USMLE Step 1 and Step 2 CK, MBBS across years and subjects, and a hybrid mode for people sitting both. BSc — chemistry, biology, physics, mathematics, computing and psychology, semester-shaped. MSc — research methods, literature critique and dissertation work. You choose once at signup; it is not a setting you flick between.
Every answer is routed across multiple providers by task and quality tier — the best available model for the job, with automatic failover. If a provider drops mid-answer, the turn moves to another lane, almost always before a word has reached you — and a retried answer is never billed twice.
Stella is built for the parts of learning that are yours alone: retrieval, explanation, drill, retention and planning. Use her alongside teaching, not instead of it — most users bring the lecture in, and let Stella make it stick.
Cancel from the dashboard. No contract, no exit interview, no dark pattern between you and the button.
Your first session is the real product.
Sign up free and work one topic properly from start to finish. Everything you touch is scored, filed, and pointed at what you got wrong — so the next session starts where this one ended.