localwriter: A LibreOffice Writer extension for local generative AI

Consider donating to support development: https://ko-fi.com/johnbalis

About

This is a LibreOffice Writer extension that enables inline generative editing with local inference. It's compatible with language models supported by text-generation-webui and ollama.

Table of Contents

• About
• Table of Contents
• Features
  • Extend Selection
  • Edit Selection
• Setup
  • LibreOffice Extension Installation
  • Backend Setup
    • text-generation-webui
    • Ollama
• Settings
• License

Features

This extension provides two powerful commands for LibreOffice Writer:

Extend Selection

Hotkey: CTRL + q

• This uses a language model to predict what comes after the selected text. There are a lot of ways to use this.
• Some example use cases for this include: writing a story or an email given a particular prompt, adding additional possible items to a grocery list, or summarizing the selected text.

Edit Selection

Hotkey: CTRL + e

• A dialog box appears to prompt the user for instructions about how to edit the selected text, then the selected text is replaced by the edited text.
• Some examples for use cases for this include changing the tone of an email, translating text to a different language, and semantically editing a scene in a story.

Setup

LibreOffice Extension Installation

1. Download the latest version of Localwriter via the releases page.
2. Open LibreOffice.
3. Navigate to Tools > Extensions.
4. Click Add and select the downloaded .oxt file.
5. Follow the on-screen instructions to install the extension.

Backend Setup

To use Localwriter, you need a backend model runner. Options include text-generation-webui and Ollama. Choose the backend that best suits your needs. Ollama is generally easier to set up. In either of these options, you will have to download and set a model.

text-generation-webui

• Installation instructions can be found here.
• Docker image available here.

After installation and model setup:

1. Enable the local OpenAI API (this ensures the API responds in a format similar to OpenAI).
2. Verify that the intended model is working (e.g., openchat3.5, suitable for 8GB VRAM setups).
3. Set the endpoint in Localwriter to localhost:5000 (or the configured port).

Ollama

• Installation instructions are available here.
• Download and use a model (gemma3 isn't bad)
• Ensure the API is enabled.
• Set the endpoint in Localwriter to localhost:11434 (or the configured port).
• Manually set the model name. (This is required for Ollama to work)

Settings

In the settings, you can configure:

• Maximum number of additional tokens for "Extend Selection."
• Maximum number of additional tokens (above the number of letters in the original selection) for "Edit Selection."
• Custom "system prompts" for both "Extend Selection" and "Edit Selection." These prompts are prepended to the selection before sending it to the language model. For example, you can use a sample of your writing to guide the model's style.

License

(See License.txt for the full license text)

Except where otherwise noted in source code, this software is provided with a MPL 2.0 license.

The code not released with an MPL2.0 license is released under the following terms. License: Creative Commons Attribution-ShareAlike 3.0 Unported License, License: The Document Foundation https://creativecommons.org/licenses/by-sa/3.0/

A large amount of code is derived from the following MPL2.0 licensed code from the Document Foundation https://gerrit.libreoffice.org/c/core/+/159938

MPL2.0

Copyright (c) 2024 John Balis