In my computing life, I prefer light-weight tools that are not distracting. I like them to do one thing, and one thing well, with only a few controls on the screen. I like to concentrate on content, usually text. I don't like anything that tries to distract me.
I am starting to put together a toolbox of my own. Part of that toolbox is a writing program called StoryPocket. It is basically a filebrowser combined with a simple text editing app. You organize your writing into directories, for example one for stories, another for gemlog posts.
Inside the directories you have either a) more directories, b) files, or c) both. Clicking on a directory goes into it. There is also a "New Scene" button that will create a text file, that you can then edit.
Clicking on a file opens up a simple text-editing window, with "Save" and "Close" buttons, which do exactly what they sound like. That's it!
There are a couple of additional details about StoryPocket. It works great as a desktop/laptop app, but it also fits on the screen of my Librem 5. So I can edit my writing from anywhere.
I sync the files with my server (Linux desktop in my house) using a script utilizing rsync commands. The laptop and phone run the script, and it syncs with my home server. I always run the script before I start editing, just to make sure everything is up-to-date.
That really is all there is to StoryPocket. I haven't shared it with the community via Git quite yet. I have a bit more editing to do, cleaning up the code. But it is a tool I am enjoying. I am writing this post in StoryPocket right now. Getting distractions out of the way helps me to just concentrate on the text. If I want to have both a story scene, and notes open, I can always run two instances of StoryPocket side-by-side. And if I need to search or anything else, I can always drop to the Terminal to run commands.
I like to use LLMs to do research for my stories. I have found myself asking LLMs questions that I use for my writing, such as "What type of trees are found in this area?", rather than jumping to the web. Having a local LLM is like having a very small set of the Internet on your local computer, with natural language search.
While I am very familiar with cloud models, and use OpenClaw personally with a cloud LLM (currently DeepSeek 4.1 Flash), what I really want is the use of local models that actually have a tool chain.
There are machines coming out that can run insanely large models now. Apple released new M5 Mac Studio machines that can support insane amounts of RAM. The M5 Ultra Mac Studio has an option coming next month (October) that will support a 512 GB RAM option. That could run a huge model locally. NVidia has their Spark computers, also able to run huge local models. Or, closer to my preferred tech stack, there is the Thelio Mega from System76, though to get a configuration powerful enough to run frontier AI models locally will cost a lot.
Machines like that being available are cool, but I am not inclined to spend tens of thousands of dollars on a computer. So I am working on being able to use smaller models instead. The most powerful local model I have installed right now is dolphin-2.9.3-mistral-nemo-12b. It is a good model, but does not have a toolchain. It is basically a chat bot. It cannot access the Internet, or the file system.
Dolphin Chat* is a C++ program that I am working on. Its purpose is to run local models and provide tool chain functionality for them. I have a large wishlist of what I want DC to do. I want it to be able to access the web, access Gemini, read and write files, remember chats across sessions, be able to execute programs, quickly switch between prompts, utilize a RAG database, and more. The list keeps growing.
The version of DC I have running right now has some of those features, mainly saving/loading chats, changing prompts quickly, and I can feed it data from the web, Gemini and files. The app handles the tool-chain, and the model just acts on the information I give it. I am very early in the project, but it is proceeding.
I am sure I will write more about this project as time goes on, and as DC gets more capable. Similar to StoryPocket, I am not yet ready to share the code. But I will when it is a little more mature.
As a developer, I like to work on and with my own tools, that I can customize as I need. I have recently shared a couple on my GitLab repo, mainly my Pure Cache app (geocaching) that runs on the Librem 5, as well as a Rosary app (also L5 compatible). I will be putting these tools in the same place when they are ready, and they will be under the GPL license. I also have a couple of browsers (a web browser, and a Gemini browser) that are also in their early stages.
* Dolphin Chat is a temporary name, chosen since I am using a Dolphin model. The name will change.