The profile and the digest

Your profile is everything langweave knows about your reading in one language. Your assistant never sees it whole; it asks for a digest — a short, plain-text summary written for it — before its first real reply in a conversation, and writes the reply from that.

What the digest says

This is a digest exactly as it is sent, for a learner a little way into Spanish with the default settings:

LANGUAGE PROFILE (format v8)

WEAVING RULES
- Replace at most 25% of content words per response.
- Introduce at most 2 new word(s) per response.
- Gloss every LEARNING word in parentheses on first use, e.g. "la ventana (window)".
- Use KNOWN words without glosses. A parenthesis showing a word's own reading — 山（やま） — is not a gloss: keep it where an entry shows one, and add none where it does not.
- Write each word exactly as it appears here. Where an entry shows two spellings separated by ／, write either one.
- If `log_exposures` is among your tools, log each word you use exactly as you wrote it, keeping the parenthesis where the entry shows one.

KNOWN WORDS (no gloss)
el edificio, el sol, la casa

LEARNING WORDS (gloss on first use)
- por favor (please)
- la ventana (window)
- la montaña (mountain)

INTRODUCE NEXT (max 2 per response)
- el río (river)

GRAMMAR IN USE
- Use present-tense verbs freely.

GRAMMAR NEXT
- ser vs estar

Section by section:

• Weaving rules — the limits. The two numbers are your settings: at most a quarter of the content words in a reply, and at most two new words per reply, by default. They are stated as limits, not targets, with an instruction to weave less when in doubt.
• Known words — written as they are, with no gloss. Up to 200 are listed (a setting); past that, the ones that match the conversation’s topic come first, then the most frequent.
• Learning words — every word you have met but do not know yet, each with its gloss, which goes in parentheses the first time the word is used in a reply.
• Introduce next — the next new words, as many as the per-reply cap, from the top of the frequency list. If the conversation has an obvious subject, it is passed along as a topic and the candidates lean toward that topic.
• Grammar — concepts you have said you know become short rules your assistant may use; concepts you are learning are named so it can lean on them. Grammar moves only when you mark it, never from exposures.

Each word is spelled the way you can read it. Spanish nouns carry their article (el edificio) where the gender reaches langweave, which it does not yet — see Spanish. Japanese words appear in kana until you know them, then with furigana, then bare — see Kanji.

When it changes

The digest is built fresh on every request; nothing is cached. Your assistant does not re-fetch it every turn — only when a tool result says your profile changed, when you mark a word, or when the topic shifts enough to be worth a new hint. A conversation that has gone on for an hour may be weaving from a digest that is a few exposures old, which is harmless: a word needs many sightings, and an extra one changes nothing.

Which language

You can be enrolled in more than one language. Each has its own profile, and one of them is active — the one woven when a call does not name a language. Your first enrollment is active; switch it in the app’s Settings, and the next request uses the new one.

There is no level and no placement test. A new profile is empty: no known words, the first two words of the frequency list waiting to be introduced. Add words in the app if you already know some.