localwriter
A LibreOffice Writer extension that adds local-inference generative AI features.
latest release: 2024-09-24 04:57:04
Description
Consider donating to support development: https://ko-fi.com/johnbalis
This is a libreoffice writer extension to allow for inline generative editing with local inference. It can be used with any language model supported by text-generation-webui.
This extension adds two powerful commands to libreoffice writer:
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.
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.
First go to tags, and download the most recent version of localwriter.oxt.
In libreoffice, Navigate to Tools, then Extensions Manager, then click Add and find the localwriter.oxt file in your filesystem. You will be directed to read the license then the extension should be installed.
You will also need to setup text-generation-webui, and enable the local openai API (note that this just means in this mode text-generation-webui will respond in the same format as open AI, all the computation still happens on your computer!). You can use whatever model you want, but I'd recommend openchat3.5 if you don't have a particular model in mind and have only 8 gigabytes of VRAM. Once the OpenAI-style API is running on localhost port 5000, localwriter should work with no additional steps. If you are using Ollama as a backend, you need to specify the model you are using in the settings.
Extend Selection is bound to ctrl + q by default, and Edit Selection is bound to ctrl + e by default.
(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
Homepage: https://github.com/balisujohn/localwriter
Repository: https://github.com/balisujohn/localwriter
Release | Description | Compatibility | Operating Systems | License | Release notes | Updated | |
---|---|---|---|---|---|---|---|
v0.0.5 | Adds max token counts and system prompts to settings. | 5.0 | Linux, Windows, macOS | MPL | v0.0.5 Release Notes ExtendSelection and EditSelection now can have max tokens to generate specified individually in settings. ExtendSelection and EditSelection now can have system prompt specified individually in settings. |
2024-09-24 04:40:14 | Download |
v0.0.4 | Fixes ExtendSelection and EditSelection for text inside of tables. Adds the localwriter logo to the oxt file. | 5.0 | Linux, Windows, macOS | MPL | # v0.0.4 Release Notes * `ExtendSelection` and `Edit Selection` now work on text inside tables. * The localwriter oxt file now has a logo! |
2024-09-02 23:53:05 | Download |
v0.0.3 | Adds an optional model setting (required for Ollama). A model key is only added to request if model is set to a nonempty string | 5.0 | Linux, Windows, macOS | MPL | v0.0.3 Release Notes Adds an optional model setting (required for Ollama). A model key is only added to request if model is set to a nonempty string. |
2024-08-01 23:48:29 | Download |
★ ★ ★ ★ ☆
Post your review
You cannot post reviews until you have logged in. Login Here.
Reviews
Marty Friedrich Sep 12, 2024, 6:55 AM (4 months ago)
★ ★ ★ ★ ☆
A really great addon! The option to use streaming for quicker responses would be great though.
(Using ollama)
John Balis Sep 24, 2024, 4:43 AM (4 months ago)
In the newest release (On Github, pending release here) You can set the max tokens per generation in settings. If you set it to a very small number, you can get something pretty similar to streaming. I am considering adding actual streaming.
No one has commented on this page yet.
RSS feed for reviews on this page | RSS feed for all reviews