Kaption wouldn't exist without Dimbreath.

// What Kaption uses this data for

Every line of dialogue Kaption recognises on your screen gets matched through a graph built from Dimbreath's data. Every NPC name we show, every quest banner, every predicted next line — it all starts there.

• →
  Dialogue graphs — Kaption predicts the next line mid-conversation so translations appear before the game finishes typing. That graph is built directly from the Talk / DialogExcelConfig data in the repos.
• →
  NPC name resolution — when the game shows a speaker name, Kaption matches it against an NPC index so the translated line knows who's talking and can pick the right pronoun and register.
• →
  Quest banners — the contextual banner above each scene (which quest you're in, what chapter) is keyed off the Quest metadata.
• →
  TextMap translation mapping — the English-to-Polish mapping that drives our translation pipeline is keyed by the TextMap hash IDs from the game, which only surface publicly because of this work.

// The repositories

// About the data

AnimeGameData and turnbasedgamedata are open community archives of the structured data inside Hoyoverse games — the raw tables the games themselves read at runtime. Talk nodes, NPC rosters, quest chapters, item lookups, and the TextMap hash-to-string dumps that every localised string keys off. They're updated each patch so tools downstream don't fall out of sync with live content.

Without this, every third-party companion tool — wiki viewers, build planners, story encyclopedias, translation overlays like Kaption — would either break on every patch or stop existing. It's slow, thankless, patch-on-patch maintenance work, and Dimbreath has been doing it for a long time.