Any book, performed by a full cast — kept true, kept yours, book after book.
▶ Hear a 90-second sample — the same passage as one narrator, then the full cast: castwright.ai/#demo.
Turn a manuscript into a finished, full-cast audiobook on your own machine — every character in their own voice, consistent across a whole series. Castwright runs locally end-to-end; nothing leaves your computer unless you opt into a cloud analyzer.
Upload .md / .txt / .epub / .pdf / .mobi / .azw3 → an analyzer
extracts characters, chapters, and per-sentence speaker tags → assign a voice to
each character → generate per-chapter audio → listen, revise, and export to M4B
or MP3.
- Full-cast performance — every character speaks in their own voice; one narrator can't be everyone.
- Series memory — a character keeps the same voice across every book in a series, even when an author renames someone mid-series. (No other tool does this for readers.)
- Designed voices — every character gets a unique voice designed from its persona, kept consistent across the series. (Cloning a voice from your own sample is the next major release.)
- Five languages — performs English, Russian, Spanish, French, and German end-to-end, with the manuscript's language detected on import. A cast never crosses languages within a book.
- Quality-checked automatically — every chapter is gated before it's assembled: acoustic checks (dead air, over-long lines, timing drift), optional word-for-word ASR transcript verification, and an opt-in speaker-fingerprint check that catches a voice drifting out of character even when the words are right. Flagged takes are surfaced on the waveform and re-recorded automatically.
- On-device by default — analysis and speech run on your machine; cloud is opt-in.
- You own the files — export standard M4B / MP3 / AAC / Opus and keep them. No lock-in.
The full user guide — installing, uploading a book, casting, generating audio, listening, exporting, and every feature area — lives on the wiki, illustrated with real screenshots.
- New here? Start at Getting Started.
- Installing from the release zip? See INSTALL.md.
- Release history: RELEASE_NOTES.md.
Castwright is source-available — not OSI open source. The code is licensed under the Functional Source License v1.1 with an Apache-2.0 future grant (FSL-1.1-ALv2, a.k.a. FSL-1.1-Apache-2.0) — see LICENSE. In short: use, modify, and share it for any purpose except building a competing product or service; two years after each release, that release's code becomes Apache-2.0. Leading with this plainly is deliberate — it bars a competing fork from day one while keeping the source fully readable.
- Name & brand — the Castwright name and identity are all rights reserved and
not part of this repository; see
docs/legal/brand-and-trademarks.md. - Model weights carry their own upstream licences (NOTICE). Coqui XTTS v2 is non-commercial (CPML) and is therefore download-on-demand, never bundled.
- Privacy & terms — the apps collect nothing and send nothing to us; the full policies live at castwright.ai/legal/privacy and castwright.ai/legal/terms. Castwright is a registered business name (ABN 52 641 247 931) in Australia.
Contributing: issues welcome; PRs by invitation for now (a DCO sign-off and a lightweight CLA apply — see CONTRIBUTING.md).