Prosody
All of the docs
  1. What Prosody is
  2. What you need
  3. Installing
  4. Signing in
  5. Adding books
  6. Syncing a book
  7. Reading and listening
  8. On your phone, and offline
  9. How the aligner works
  10. When something is wrong
  11. Updating and backing up
  12. Where books come from
  13. Words

What you need

The host hardware, supporting software, build tools and book formats required to run Prosody.

Prosody runs on a single Linux machine with no GPU. Alignment runs on the CPU, and the rest of the stack is light enough to share the host with other services.

The host

RequirementWhy it is needed
Linux with Docker (swarm mode)Prosody runs Postgres, the model runner, the Api and the press as Docker swarm services, alongside a Caddy relay container. A single-node swarm is enough.
Multi-core CPUThe speech model runs on the CPU using one thread per physical core by default. More cores shorten the initial listen; no GPU is used.
Local disk for reading copiesThe press writes reading copies to a local folder. Each synced book takes roughly twice the size of its audiobook (an audio copy plus an EPUB containing the audio, alongside the saved speech emissions).
About 3.2 GB for the press imageThe press container image (3.15 GB) bundles ffmpeg, poppler and both speech models, downloaded once during the first build.

Syncing speed

Listening to an audiobook is the slow step, and it happens only once per audio file. Measured across five real books (72 hours of audio in total) on an 8-core CPU-only Linux host:

Audiobook lengthTime to listen
9.9 hours33 minutes
11.5 hours32 minutes
13.5 hours62 minutes
16.2 hours48 minutes
20.9 hours56 minutes

Across those five books, listening ran 13 to 23 times faster than real time (19 times overall). Once listening finishes, aligning the text, building the reading copy and running the spot check takes under 35 seconds. Because the speech model's output is saved beside the reading copy, re-aligning the same audio skips listening altogether.

Networking and sign-in

RequirementWhy it is needed
Tailscale on the hostThe Api queries the host's Tailscale daemon over /var/run/tailscale/tailscaled.sock to identify each device.
Tailscale on every reading deviceOnly your own untagged Tailscale devices are admitted.
Caddy on the hostCaddy terminates TLS, checks every request via forward_auth, and proxies to the relay on loopback.
A domain name for the appFor example prosody.example.com, pointing to the host's tailnet address, with a TLS certificate Caddy can serve.

Tools for installing and updating

Run these on the host where you deploy:

ToolVersion / notes
Node.js26.11.0 or newer
pnpm12 (corepack enable activates the pinned version)
BunRequired by the Alchemy deploy tool

Your book files

  • Ebooks: DRM-free .epub files.

  • Audiobooks: DRM-free .m4b files encoded in AAC-LC, one file per book.

  • PDFs: Plain .pdf files, displayed one page at a time.

See Where books come from for DRM-free sources, and Adding books for the folder layout.