Models keep getting smarter.
What compounds is how you brief them, what you give them, and what you trust.
That's the Human Layer.
Clarity.
Scope the right question before you spend a token.
Frontier models can write anything — including the wrong thing, beautifully. Clarity is the habit of narrowing the brief: who is this for, what does done look like, what's out of scope. Keypra trains it with instant scoring, structural decomposition and a coaching loop that nudges, never solves.
What you'll feel
- Spot ambiguity before the model does
- Decompose a fuzzy ask into named blocks
- Compress without losing the brief
What good looks like
Before
"Summarise this contract."
After
"Review the attached supplier agreement under German BGB §305–310. Identify (1) liability caps and triggers, (2) termination clauses with notice periods, (3) GDPR processing obligations. Flag conflicts with our Procurement Policy [Context Card]. Output as a table with risk ratings."
Context.
Give the model what it cannot infer — once, and reuse it forever.
The fastest way to ruin a great model is to under-brief it. Context turns your domain expertise into reusable assets — Wings and Rooms in your Palace, Context Cards, AI Personas and full Skill Bundles — so every prompt starts where your last one left off, not from scratch.
What you'll feel
- Find anything by Wing, not by folder
- Reuse a Card across ten prompts in one click
- Ship a Skill Bundle the team can install
Your Palace, at a glance
Legal Wing
12 rooms
Finance Wing
9 rooms
Sales Wing
14 rooms
HR Wing
6 rooms
Confidence.
Know what to trust before it leaves your screen.
Frontier models fail invisibly. Confidence is the habit of inspecting outputs against your own quality bar — structured checks, retention prompts that confirm the practice stuck, and a Competence Dashboard that members own outright. Employers see aggregate signals only, by Article 4(b) design.
What you'll feel
- Catch silent failures before the reader does
- See your three-pillar profile sharpen over weeks
- Prove the training landed, not just that it happened
What good looks like
The Classroom layer
Learn live. Or learn at your own pace.
Live — Classroom & Rafy
Lecturer-led sessions with synced slides, real-time exercises and Rafy, our co-creator who keeps every learner on the same step.
- · Synced slide playback for cohorts
- · Real-time exercises and peer review
- · Rafy keeps the room together, never ahead
Async — Curriculum & Academy
Piece-by-Piece lessons, Weekly Quests and an XP path that compounds. For organisations, the Learning Academy adds citizen-lecturers and a managed learning pool.
- · Piece-by-Piece curriculum blocks
- · Weekly Quests and Compression Challenges
- · Learning Academy for enterprise cohorts
The Organisation layer
Built for European enterprises.
Compliance is a feature, not a tax. Keypra ships the controls your IT and Legal teams expect — without slowing down the people doing the work.
BYOK & EU-Model policy
Bring your own model keys; restrict the workspace to EU-hosted models when policy demands it.
Aggregate-only analytics
Article 4(b) safe by design. Members own their detail; managers see anonymised cohort signals.
Audit trail
Exportable evidence of training, prompts and BYOK lifecycle — ready for internal and external audits.
Org Library & Sharing
Centralise corporate IP — Cards, Personas, Skill Bundles — with deep-cloning across teams.
Enterprise customers receive a dedicated contract and named support — no consumer billing flows.
Try a rep
Paste a prompt you used this week.
We'll open it in the Practice Lab and walk you through scoring, decomposition and a coaching loop. No sign-up required to look around.
Common questions
Honest answers.
Is this another prompt library?+
No. Libraries hand you fish; Keypra is the gym where you learn to fish. Every surface — Lab, Composer, Palace, Skill Forge — exists to build a habit, not to hand you a one-off output.
Do we need IT to roll this out?+
No. Keypra is web-only — no desktop install, no plugin. Enterprise teams can add SSO (Google today; Microsoft Entra ID and SAML on the roadmap) and bring their own model keys (BYOK) when they're ready.
How does this map to the EU AI Act?+
Article 4 makes AI literacy a duty for organisations using AI. Keypra delivers practice, evidence and an exportable audit trail. Article 4(b) safeguards are built in: no per-person AI-mastery classifications are ever shown to employers — aggregate-only by design.
Will this become obsolete when models improve?+
The opposite. The Human Layer compounds: as models get sharper, the people who brief them clearly, give them the right context and verify their output pull further ahead. That's what we train.