I wrote previously about my attempt to design, learn and use a better keyboard.
- Towards a better keyboard
- Screw it, I’m learning QGMLWY
- Layouts review: Dvorak vs Colemak vs Carpalx vs Workman
- Special characters layout
All those ideas are getting incorporated into the keybest repository. The design is mostly complete, I still need to learn the special characters layout as well as some macros.
The next step is Dial.
Keybest is about a better typing method. It is about being able to type efficiently.
Dial is about a better input method. It is about being able manipulate data (text mostly) in and out of programs efficiently.
What does it mean?
Dial Is A LSTM. (I plan to implement it as a LSTM.)
I like the name because it is descriptive (it really says what the thing is). It is recursive. LSTM being a type of recurrent neural nets justifies the recursion. And also because it is about dialing stuff.
What does it do?
The basic idea is having one text box to rule them all.
It will be an Emacs buffer + clipboard manager + LSTM. You can also view it as an AI typing assistant.
With Dial, most text input is typed into this box and then copy/pasted around. A key requirement is that it should play nice with a clipboard manager. And because they are no good clipboard managers around, I think Dial will double as a clipboard manager. (I’ll do a more extensive search, but so far, Flycut seems the only one. I’m using it and it is nice, but it is way to limited).
List of features:
- Text editor (probably an emacs buffer running in a terminal. Will see for the details)
- Buffer management (merge buffer, new buffer, delete buffer, export buffer to clipboard, import buffer from clipboard). The active buffer probably will be the clipboard content, with merging macros.
- Deep integration with clipboard manager
- Source code highlighting (including markup languages, markdown, and programming languages)
- Source code completion
- Spell checking
- LSTM Completion (killer feature imo): completes based on every personal input so far, as well as a big index computed from internet content
Nice to have:
- Instant launch (should be really fast to open up. Ideally less than 70-100ms)
- Automatic detection of language (french, english, html, markdown, css, c, ruby, erb, whatever)
- Support for other formats than text (files, images and video).
- Image edition would be nice (although monosnap already does this, and it is a lot of work)
- Internet completion (stackoverflow/google/wikipedia?). I’ll have to experiment with this. Probably useful with an explicit command to trigger this kind of external completion