iOS 1.9November 10, 2026

Story Atlas, word-level highlight, and a sharper Codex

The Codex Atlas charts your book's world on an interactive map, narration can mark each spoken word, and a deep entity-resolution pass makes the companion noticeably more accurate.

StoryCodex 1.9 opens the Codex Atlas to release builds, adds an opt-in word highlight for narration, brings a new neural voice engine, and carries the largest companion accuracy pass yet, plus a reliability sweep across analysis memory, playback, and billing.

Meet the Story Atlas

A new Atlas mode on the Codex graph screen charts the book's world on an interactive map. Places sort into their regions and realms, named seas get proper extents, journeys draw inside route corridors instead of cutting across the world, and large worlds stay readable through realm tints, depth-banded labels, and merged place chips that zoom on tap.

Atlas sits between the Journey and Relationships modes. Free users get the Pro preview over the map; Pro, credit purchases, and managed cloud unlock it in full.

Follow every word

A new optional setting, Highlight the spoken word, marks each word inside the narration sentence band. Find it in the reader's appearance sheet; it is off by default, and timing is estimated for on-device voices.

Chatterbox neural voices

A fifth on-device narration engine joins the lineup: Chatterbox brings six expressive English voices powered by a neural model. The download is roughly 675 MB and speaks at a fixed studio rate, and its input is normalized so headings like Chapter VIII read aloud as numbers instead of letters.

A parked system voice can no longer stall playback either. A request that wedges during the utterance handoff now times out after ten seconds and rebinds instead of hanging the session.

A sharper Codex

The entity-resolution pass now repairs bad renames, record takeovers, and mistaken merges. A family-token guard stops a minor character sharing a surname from hijacking a major record, and a one-time audit heals already-damaged codices on first launch after updating, restoring taken-over names and splitting intruders into their own entries.

A new per-session High precision codex option runs a stricter pass for books that need it, taking longer and costing more for sharper entries. Profile images that lost their entity are cataloged instead of silently orphaned, and EPUB synopsis and works-by pages now classify as front matter so they stop polluting extractions.

Analysis without the memory spike

The EmbeddingGemma model now releases on an idle timer between chapter passes instead of reloading its roughly 315 MB natively for every chapter. Bulk codex runs no longer balloon toward the low-memory kill threshold that could take the foreground app down mid-run.

In this release

More improvements

  • Codex enable surfaces now recommend the embedding model download so deduplication and related-history lookup have what they need
  • Turning the companion on mid-chapter now starts its analysis right away instead of waiting for the next tap
  • The settings voice row labels an installed engine's out-of-catalog speaker correctly

Fixes and improvements

Bug fixes

  • Interrupted credit purchases recover: sessions retry sign-in before reserving, an already-owned product re-queries its pending grant, and account linking only takes the empty-wallet fast path when the server confirms it is empty
  • The system voice engine no longer stalls playback on a parked utterance; the handoff now times out and rebinds

Open the Codex graph on a book with companion summaries and switch to Atlas, then turn on Highlight the spoken word from the reader's appearance sheet to try the new narration mark.

Keep the story clear

Try StoryCodex for free.

Read, listen, and remember any long story with a private, spoiler-safe story memory that lives on your device.