Why not just use ChatGPT?
A fair question. We're not going to bash other tools — we're going to be honest about what they cannot do for a university application, and what you would have to build yourself to match.
Capability
UniOffer
DIY / generic AI / open-source template
Admissions Readiness Engine
Calibrated to each university's published Common Data Set + UCAS/HESA. No probability numbers shown — qualitative bands tied to real selection signals.
An LLM guess. No published-admit-data anchor. The model will confidently invent a percentage.
School database
US Top 50 + UK Top 20 + QS World Top 100 with verified admission data where available. International schools clearly tagged as QS-rank estimates, never dressed up as precise.
Whatever the model remembered from training data, plus whatever it can search live. No guarantee a number is current or accurate.
Student DNA / persona
8-dimension candidate profile built once from CEF, then anchored across every agent output and every essay revision so your voice and narrative stay coherent.
You re-paste your background into every chat. Persona drifts every session because the model has no memory of who you are.
Cross-session memory
Your CEF, essay versions, school list, advisor sessions, readiness signals — all persist. Walk away for two weeks, come back, the agent picks up exactly where you left off.
ChatGPT memory is shallow and tool-scoped. Re-paste everything each session.
Agent specialisation
12 specialist agents — positioning, school matcher, essay coach, interview, test prep, activities, leadership narrative, capstone research, LoR, scholarship finder, what-if planner, A0 orchestrator. Each carries the same persona context.
One general-purpose model. Every prompt is a fresh start; no orchestration between tasks.
Honest data invariant
Codified: no fabricated stats, no "guaranteed admission", no fake review counts (see docs/AUDIT-MEMO.md). Every value on screen traces to a real source.
Generic LLMs hallucinate acceptance rates, deadlines, and faculty names on demand. You learn that the hard way.
Cozie wellbeing companion
First-class wellbeing presence with pre-LLM crisis-keyword filter routing to Samaritans (UK) / 988 (US). Schema-isolated from admissions data per UK Online Safety Act 2023 Article 9 stance.
A model that will gamely roleplay as a therapist with no safety net. No crisis-detection escalation, no jurisdictional hotline routing.
Compliance scope
GDPR (EU/UK), COPPA (US under-13), PIPL (CN), FERPA, UK Online Safety Act 2023, IECA & NACAC professional standards — all designed in.
A single ToS that disclaims everything. Your data crosses providers as you switch tools.
Expert-in-the-loop
When the loop genuinely needs a human — former admissions officers at Oxbridge / Ivy / Russell Group — you can pull one in without leaving the workspace.
You hire a separate consultant on a separate platform and re-explain your situation from scratch.
Setup cost
Sign up, complete CEF (≈45 min). The whole stack — agents, data, compliance, memory — is on from day one.
A weekend to wire up a LangChain template, ongoing maintenance, and you still don't have the data or the safeguarding stack.
We are not claiming generic AI is bad. It is excellent at one-off drafting and quick research. We are claiming that a university application is a 12-month, multi-document, compliance-sensitive project for a 15–18 year old — and that for that project, you want a system that remembers you, carries verified data, and was designed for minors from day one. That is what we built.