Home Reader turns source text into a listening script with a local Ollama chat model. On macOS it can also narrate the script into a WAV file with the built-in say and afconvert tools. It includes a local browser interface, a command-line interface, and a persistent library of saved readings.
Home Reader is a local application, not a hosted cloud demo. It has no runtime dependencies, accounts, analytics, or build step.
- Node.js 22 or newer
- Ollama running with at least one installed chat model
- macOS for audio generation; Linux and Windows users can generate text with the CLI's
--no-audiooption
Start Ollama if it is not already running:
ollama serveHome Reader prefers qwen3:8b when it is installed. Pull it only if you need it:
ollama pull qwen3:8bYou can instead select a smaller installed chat model if that is a better fit for your machine.
No npm install is required.
git clone https://github.com/Metatransformer/home-reader.git
cd home-reader
npm startOpen http://127.0.0.1:3210. Paste up to 12,000 characters, select an installed model, and choose one of three modes:
- Quick: the main idea, key facts, and important caveats
- Explain: a plain-English explanation with context
- Deep: a detailed walkthrough
The browser library persists saved readings across server restarts. Each reading may include the original source, generated listening script, WAV audio, and metadata in its folder under outputs/.
Read from a file:
node cli.mjs --file article.txt --type deep --model qwen3:8bRead from standard input:
cat article.txt | node cli.mjs --type explainGenerate text without macOS audio tools:
node cli.mjs --file article.txt --type quick --no-audioRun node cli.mjs --help for all options. The same 12,000-character input limit applies to the browser and CLI.
Readings are stored in outputs/<reading-id>/. Home Reader saves source.txt and metadata.json first, then adds script.txt and audio.wav as those steps succeed. If generation or narration fails, completed partial results remain available and the metadata records the error.
The original source is saved exactly as submitted. Generated summaries and explanations can omit or misstate details, so use the original as the source of truth.
OLLAMA_URL defaults to http://127.0.0.1:11434. If you override it with a remote Ollama endpoint, Home Reader sends the submitted source text to that endpoint.
npm run check
npm testThe code is available under the MIT License. The license does not cover Ollama models or the user's source and generated content; those remain subject to their own applicable terms and licenses.