Tutors that remember each learner and adapt as they grow.
Per-user sub-brains track each student's progress and gaps; the coherence gate keeps the teaching approach steady.
A tutor that remembers each learner's progress and gaps, holds a consistent teaching method, and adapts to the individual as they grow.
Good tutoring is cumulative — it depends on remembering what a student already knows, where they struggle, and how they learn best. A stateless model resets every session, so the tutor keeps re-teaching basics and can't adapt to the individual over time.
What changes with a memory layer.
- Progress
Re-teaches the basics every session.
- Method
Teaching approach swings session to session.
- Adaptation
The same lesson for everyone.
- Privacy
Student data sits in a shared context.
- Progress
Builds on what each student already knows.
- Method
A coherence gate keeps pedagogy steady.
- Adaptation
A brain per learner that sharpens over time.
- Privacy
Per-learner isolation; on-prem for institutions.
The same loop — three calls, your model in the middle.
prepare
Builds the brief — memory + written identity + coherence gate. No LLM call.
your model
You call your own model — your provider, your key. Khwan never touches it.
record
Persists the turn and learns from it — so the next prepare is sharper.
Every run compounds. The brief gets tighter, cost drops, and answers sharpen — the memory that thinks, not just recalls.
A brain per learner
Per-user sub-brains keep each student's progress, misconceptions, and pace persistent across sessions — so the tutor adapts instead of restarting.
Consistent pedagogy
A written teaching approach becomes the coherence gate, keeping tone, difficulty, and method steady rather than swinging session to session.
It compounds
Synthesis turns each learner's history into a sharper model of how to teach them next — the tutor genuinely improves for that student over time.
Outcomes, not orchestration.
- Lessons that build on what each student already knows
- Consistent teaching method per learner
- Adaptation that improves with every session
- Private by design — on-prem available for institutions
Edtech products, tutoring platforms, and training tools.
How Education wires Khwan — three calls, your model in the middle
import { Khwan } from "@khwan/client";
// A brain per learner — tracks progress + gaps across sessions.
const kw = new Khwan({
apiKey: process.env.KHWAN_API_KEY!,
userId: student.id, // isolated sub-brain per learner
});
// 1 — Recalls what this student knows and where they struggle.
const turn = await kw.prepare(question);
// Coherence gate: keep the teaching method steady.
if (!turn.allowed) return turn.reason;
// 2 — Your model teaches, adapted to this learner.
const answer = await yourModel(turn.messages);
// 3 — Record it; the tutor sharpens for this student over time.
await kw.record(turn, answer);