prose-ai

A local, offline prose editor powered by a language model running on your own machine. Paste or write text, pick an editorial mode, and get inline tracked changes you can accept or reject one by one. No cloud, no subscription, no data leaving your machine.

Modes: Proofread · Rewrite · Formalize · Concise

Highlight any word or phrase to get contextual synonyms via a bubble menu.

Note

Editorial defaults (Oxford commas, em-dash handling, sentence capitalization, etc.) are defined in src/llm/prompt.js in the SHARED_RULES section. Adjust them to match your own style guide.

---

What runs it

• Ollama — runs the language model locally
• gemma3:4b — Google's 4B model, fast on Apple Silicon and modern CPUs Model and chunking configuration lives in src/llm/config.js.
• Bun — JS runtime and package manager
• Vite — dev server and bundler
• Tiptap — rich text editor (MIT, open source)

---

Install Ollama

macOS (Homebrew)

brew install ollama

macOS (direct)

Download from ollama.com/download and run the installer.

Linux

curl -fsSL https://ollama.com/install.sh | sh

Windows

Download the installer from ollama.com/download.

---

Start Ollama and pull the model

# start the ollama server
ollama serve &> /tmp/ollama.log &

# pull the model (~3.3 GB)
ollama pull gemma3:4b

# warm it into memory (optional — skips cold-load delay on first use)
ollama run gemma3:4b "" &> /dev/null &

Tip

If you use the Ollama macOS menu bar app, it handles ollama serve automatically, so you can skip that step.

---

Install and run prose-ai

# install dependencies
bun i

# build
bun run build

# start the dev server
bun run dev

Open http://localhost:5173.

---

Credits & License

by xero & claude code

CC0 1.0 Universal — public domain, no rights reserved.