Skip to content
The public beta is open — download Castwright free.
Castwright

Free, every book, the full cast — yours to keep.

Every character. Their own voice. Every book in the series.

Any book, performed by a full cast — kept true, kept yours, book after book. It runs on a machine you may already own — a gaming PC or laptop with a 6 GB graphics card, or any Apple Silicon Mac — and your book never leaves the house.

No account, no upload, no catch. The source is on GitHub for you to read; the voices are Kokoro, Coqui XTTS and Qwen3-TTS; the whole performance runs on your own machine.

If you can game on it, you can cast on it.

The work falls on a graphics card you may already own — 6 GB gets you started and 8 GB is the sweet spot, the kind in a mid-range gaming laptop, or any Apple Silicon Mac. No GPU? It still runs on the processor alone — slower, and we'll say so plainly rather than pretend otherwise. Check what you need →

Listen first

One narrator, or a whole cast.

Hear it — The forge, and the dragon at the door

The Coalfall Commission

0:00 / 0:00

Now playing: Full cast — every character in their own voice

The cast you’ll hear

  • Wren · “SparrowGirl, 13 — bright, quick
  • Master OduvanMan, 70 — gravel and patience
  • MaerinWoman, 40s — clipped, brisk
  • CoalfallThe dragon — low, slow, ancient
  • NarratorWarm, measured, unhurried

Same passage, two ways. One narrator, then every character in their own voice.

Five languages to hear here. Chinese and Japanese are performed too — those clips are still in the studio.

How it works

From a book to a performance.

  1. 1

    Drop in a book — EPUB, PDF, or paste a chapter.

  2. 2

    It finds the cast — every character, detected and attributed.

  3. 3

    Give them voices — assign or design a distinct voice each.

  4. 4

    Hear the performance — a full-cast audiobook, yours to keep.

The app

A finished tool, not a script.

Castwright listen view

Listen — chapter by chapter, full cast or narrator

Castwright cast confirmation view

Cast — every character, voice assigned and previewed

Castwright generation view

Generate — watch each line render in real time

Castwright library view

Library — your series, organised and always playable

Four reasons it sounds right.

A full cast

One narrator can't be everyone. The apprentice sounds thirteen; the swordsmith sounds seventy and a forge.

No one else does this

Series-consistent

Book two should sound like book one. Castwright remembers your cast — even when an author renames someone mid-series.

Your own voice In development

Read a bedtime story in your own voice, even when you're away — or let your kid be the hero. Your voices, with your permission, will stay on your machine. Coming soon.

Yours to keep

Rendered once, on hardware you own. No meter, no monthly fee, no server that can take your library away. Your book never leaves the house.

What no one else does

A library that remembers.

In The Coalfall Commission, the dragon's apprentice is called Wren by everyone except old Master Oduvan, who knew her when she was small and still calls her Sparrow. Same girl. Two names. One voice.

Castwright resolves that automatically — matching characters across books by identity, not just by what a given chapter happens to call them. And when an author renames a character between books (it happens more than you'd think), Castwright keeps the voice consistent even then.

Without series memory, the same character gets a new voice every time they cross a book boundary. The performance fractures. Castwright keeps it whole.

Full-cast generation isn't unique anymore. Series memory is.

Castwright's cast view for Saltgrave — the series voice library on the right showing voices carried from earlier Hollow Tide books, a voice-drift notice, and duplicate-character merge prompts
The series-memory reveal — 'Six books in, and not a voice has changed' — five carried voices (Narrator, Insp. Cray, Dr. Wren, Constance Vale, Magistrate Cross), each shown present across all six books of The Hollow Tide

The proof, on your own shelf: how many voices have carried, across how many books.

The whole shelf

Your whole library, in one place.

Every series on one shelf — what's in progress, what's finished, and where you left off in each. Covers, runtimes, and the cast behind every book: all yours, all on your own disk.

A full Castwright library — Continue listening, multiple series with per-book chapter and runtime stats, and an Add another book card

The road ahead

What's coming next.

Everything below is on the bench today — some of it lands free for everyone, some joins the Cast Pass.

Your voice, and your family's

In development Cast Pass

Lend the narrator your voice — recorded with consent, kept at home, never uploaded. Your first voice will be free; family voices join the Cast Pass.

More voice engines

Planned Free

Kokoro, Qwen and Coqui aren't the end of it. Two expressive engines are on the bench for bigger cards — IndexTTS-2, tuned for per-line emotion from one designed voice (likely a 12 GB card), and Fish Audio S2-Pro for the 16 GB tier, chasing the finest fidelity. They land as bigger-card support matures.

Three more languages

Planned Free

Castwright performs in English, Russian, Spanish, French, German, Chinese and Japanese today. The Qwen3-TTS engine underneath speaks three more — Korean, Italian and Portuguese — and we're bringing them to the cast, free.

One voice, all the way through

Planned Free

Series memory already gives a returning character the same voice, book after book. Next, the quality gate's ear reaches across the whole series — checking each character still sounds like themselves from the first book to the latest, not just that they were handed the right voice. Whether that same check can reach across languages is a harder, unproven question we're still exploring.

Bring your AI assistant

Planned Cast Pass

Tell Claude — or any assistant that speaks MCP — "make me the audiobook", and it drives Castwright end to end: import, cast, generate, deliver to your phone.

Proof, not promises.

Quality

Every line is acoustically checked before a chapter is assembled, and can be verified word for word against what was actually said. The plainly broken ones never reach your ears — and every finished book carries a visible, exportable receipt of exactly what was checked.

Patience

Start it and walk away. Every heavy step — analysis, voice design, generation — runs in a queue that retries the rough patches on its own: a model hiccup or a GPU stumble doesn't sink the job, it picks back up. An activity log tells you exactly what happened while you were gone.

Economics

You pay the electricity, once per book — against ~$11/month clouds that won't let you export, and cloud creation that runs to ~$200 per book by the vendors' own published rates (June 2026).

Permanence

When Play.ht was acquired in 2025, its users' libraries and voice clones were deleted at sunset. Your renders are files on your disk. Nobody can take them away.

Versus the alternatives

Honest about the trade-offs.

  • Clouds make stunning voices — they also meter them by the hour and keep your audio.
  • Open-source tools are kin; today they cast within a single book, not across a series.
  • Audible's human full-cast productions are the benchmark we honour.
See the full comparison →

Pricing

Free. The library is seven dollars.

Castwright is free — every book, the full cast, yours to keep. At launch, the Cast Pass is seven dollars, once. Free buys you a perfect book; the Cast Pass gives your library a memory.

See what's free →

Questions

The honest answers.

Do I need a GPU?
A gaming PC or gaming laptop with a 6 GB NVIDIA GPU gets you started, and 8 GB is the sweet spot — that's the benchmarked path, generating at roughly real-time. Apple-silicon Macs are supported too: the engines run on the Mac's GPU (or CPU), just slower. No GPU at all works, but it's slow enough that we'd call it a fallback, not a plan. AMD GPUs are an experimental preview: Qwen and Coqui run on ROCm and Kokoro on the CPU, with an automatic CPU fallback if the ROCm install doesn't take — a ROCm-supported card and a recent driver are needed. Analysis — the character-detection step — runs on a local CPU model or the free Gemini tier either way, so the GPU stays free for speech.
Is my data private?
Yes. Your books, voices, and library stay on your machine, and synthesis runs entirely locally once models are installed — the generated audio never leaves the house. Working out who speaks each line reads your book locally too, by default. There is an optional cloud analyser (Gemini free tier) that sends chapter text: if you have set a cloud key, it steps in when the local model isn't running. One setting stops it doing that, for good.
What does it cost?
Nothing, during the public beta: it's free for every book — the full cast, series memory included, yours to keep. At launch, the Cast Pass — seven dollars, once — unlocks the cross-book series memory and the companion app. No subscription, no metering. Free buys you a perfect book; the Cast Pass gives your library a memory. See /pricing for the breakdown.
What formats can I import and export?
Import: EPUB, PDF, or paste plain text directly. Export: M4B (chapter-marked audiobook, works in every major player), MP3 (flat audio, maximum compatibility), and SRT or VTT captions — line, sentence, or word level, for the whole book or a single chapter, handy for a demo clip or reading along. Your files land on your drive — no upload required.
All questions →

Any book, performed by a full cast — kept true, kept yours, book after book.