Immersion works, and you already live online.
Polyglot Cards moves your feed, your AI, and your flashcards
into your target language, so you absorb it by exposure.
Learning is training recall — and recall has directions. Each direction is its own memory, so one entry becomes three cards, each on its own schedule.
Immersion is a choice now. Your videos, your podcasts, your AI — all of it can happen in the language you are learning.
Every podcast you already listen to ships audio tracks in other languages.
placeholder · founder videoConnect ChatGPT and have spoken conversations that use only the words you know.
Your knowledge is a network. "Dog" is not a fact filed somewhere — it is a pattern: four legs, furry, barks, animal firing together. On day one, a card hops in through what you already have: its picture finds the dogs you have seen, собака finds the English word. That is how it should start. But every review, and every picture, voice, and sentence on your cards, grows direct pathways — until собака fires the pattern itself, no English in the loop. FSRS times the reviews that grow them.
One sentence in, fifty cards out — each arriving with its audio already generated.
The agent writes pronunciation, not just text — stress for Russian, tones for Chinese, pitch accent for Japanese.
Pairs a native speaker hears as two different words can reach your ear as one — your language never needed the split. Play one, prove which you heard.
Claude reads your review history, tells you where you slip, and builds the review for exactly that.
Polyglot Cards speaks MCP, so any Claude conversation can build and tend your decks. One address, pasted once.
In claude.ai, open Settings → Connectors and choose Add custom connector.
Paste this address and save:
https://ifucjgfizrpoesfwenad.supabase.co/functions/v1/mcp
Click Connect and sign in with your Polyglot Cards account. That login is the whole permission model — Claude only ever sees your own decks, and revoking the connector cuts it off instantly.
Then just talk: “make me a deck of the hundred most common Russian verbs” — cards, audio, and images arrive in your review queue, held for your approval.
You will look at this a few hundred times a month. Hover anything.
Ask in one sentence; an agent writes the cards, files them, and records the audio.
Neural voices tuned per language get stress, palatalization, and tone right — no hunting mp3 files, ever.
A memory model trained on billions of reviews — each card returns at the exact moment before you would forget it. The closest thing to downloading vocabulary into long-term storage.
Pairs a native speaker hears as different can sound identical to you — the trainer plays one and your ear proves which it heard.
Recall, meaning recall, cloze, relearn — the same word is tested from every direction until it is yours.
Talk with AI that stays inside the words you know — understanding everything you hear is what makes conversation practice work.
The AI reads your review history, finds where you are falling behind, and builds the lesson that catches you up.
A short listen built from the words you already know, new whenever you ask.
soonTwo findings have held for decades: you acquire a language from input you almost understand, and you keep what returns on a widening schedule.
Catch a word just before it slips and it rebounds higher — stronger memory, slower decay, longer gap. The scheduler finds that edge for every card; ten minutes a day holds a vocabulary of thousands.
Acquisition happens when you understand messages pitched slightly above your level. Not drilling, not grammar tables. Understanding.
Hundreds of studies pooled into one answer: spaced review beats massed review, reliably and by a lot. The gap is the mechanism.
Both grew into the same modern scheduler — FSRS, which returns every card at the last moment it still counts.