Collect
Collect RSS and Atom feeds, Gmail newsletters, Wallabag saves, and media sources you want included in the digest.
Open-source and self-hosted
Ghostwriter runs locally and builds digests from feeds, saved articles, and transcripts.
Run it on your own hardware with Docker. Ghostwriter produces EPUB digests, supports local transcription with whisper.cpp, and can optionally generate podcast episodes from the same material.
Workflow
How Ghostwriter processes content
Ghostwriter collects sources, prepares them for reading, and produces files you can keep and use locally.
Extract
Extract readable article text from web pages and preserve enough structure to produce useful EPUB output.
Shape
Choose whether a source should stay full-text or be summarized into a shorter briefing.
Publish
Build an EPUB digest, and optionally generate a podcast episode and private feed from the same material.
Features
Current capabilities
These are features already available in the current Ghostwriter codebase.
Self-hosted
Run it locally
Ghostwriter is meant to be run on your own machine, NAS, server, or home lab with Docker. It does not require a hosted Ghostwriter service.
Inputs
RSS, newsletters, Wallabag, and media sources
The backend can combine web feeds, newsletter labels, saved articles, and transcribed media into one digest pipeline.
Reading modes
Full text or summarized briefings
Each source can stay close to the original text or be condensed, depending on how you want to read it later.
Output
EPUB digests for offline reading
Generated digests are compiled into EPUB files that work well on phones, tablets, and e-readers.
Audio
Podcast episodes and feeds
Ghostwriter can turn digest content into narrated podcast episodes and publish them through a standard RSS podcast feed.
Transcription
YouTube and podcast transcripts
Media items can be transcribed and included in the digest, with support for local whisper.cpp transcription and API-based fallback.
Self-Hosting
Runs locally, including smaller home-lab setups
Ghostwriter is intended for self-hosting. That includes ordinary Docker setups, ARM devices, and Raspberry Pi deployments where keeping processing local matters.
Local first
No Ghostwriter cloud required
You can run the service locally with Docker and keep feeds, schedules, tokens, and generated files in your own environment.
Raspberry Pi
Works on ARM and Raspberry Pi setups
The project documents a whisper.cpp binary override for ARM and Raspberry Pi environments so transcription can stay local on smaller hardware.
Transcription
Local whisper.cpp support
Ghostwriter can use bundled or user-provided whisper.cpp binaries for local audio transcription, with OpenAI Whisper API available as an alternative.
Screenshots
Web interface
These screenshots come from the current Ghostwriter web UI.
Dashboard
Overview of service status, recent activity, and generated digests.
Feeds
Manage feed sources and decide what should be part of the digest.
Digests
Review generated EPUBs and track previous digest runs.
Settings
Configure schedules, providers, transcription, and API tokens.
Quick Start
Run it locally with Docker
The default setup is local and self-hosted. Cloud providers are optional, and local transcription is supported.
1. Configure environment
Copy the example environment file and choose the AI and transcription setup you want to use.
2. Start the stack
Run the Docker compose command, verify the health endpoint, then open the local web UI.
3. Create your first user
On first run, create the admin account and generate API tokens for any clients you want to connect.
From the Ghostwriter directory
cp .env.example .env
# choose an AI provider and transcription settings
docker compose up -d
curl http://localhost:8080/healthAdjust AI provider and transcription settings in the environment file before starting the stack.
Source and setup
The project is available in the public repository.
Start with the Ghostwriter repository, run it locally, and adapt the setup to your own hardware if needed. Epilogue can connect to Ghostwriter, but it is optional.