I’m sharing this project because I thought such a utility might be useful to someone. I decided to check if local models (SLMs) are capable eniugh for most routine tasks. It is incredibly simple, but the logic of AI Skills is not limited by anything - anything the Python language can do, this product can do. And over time, as local AI becomes even smarter, it will become even more useful.
MacOS already has the Spotlight service. What I created is a kind of analogue but for launching local AI Skills.
Technical part:
Python + Tkinter (a bar summoned by the hotkey ctrl + alt) + Ollama (qwen2.5:3b). I specifically didn't use any other heavy libraries. The project turned out very lightweight. One Click install and start .bat file.
Skill Logic:
In the root of the project, there is a /skills folder. Each .py file is a script. 1 .py file = 1 AI Skill (Command, Script, etc.). This means that skills can be shared. Like Mods in Minecraft.
Why am I convinced that someone needs this?
News recently came out about skills inside the Chrome browser. And also OpenAI's purchase of Sky for Codex. All this news is somehow related to skills and agents. And 100% everyone will eventually need an absolutely free analogue. One with which you can process even thousands of lines of text without paying a penny.
How does it work?
Copy the text you need to work with. Press the hotkey (ctrl + alt). And select the desired script (command). The script pulls the text from the clipboard and executes its logic, then returns it directly to the interface without breaking the user's workflow (I attached a GIF demonstration of the project's operation on GitHub). That is, you won't have to use a browser or make other extra movements. All you need is Ollama already installed.
About the Skills
I created several very simple skills that I needed most often.
List:
Fix Code - you feed it broken code, you get fixed code
Improve Text - rewrites text cleaner and more professionally
JSON Format - fixes and formats messy JSON
camelCase - text to camelCase
snake_case - text to snake_case
Slugify - text to url-slug
Current project status:
Right now the project is very raw and primitive, especially the skills and their logic (literally 10-20 lines of Python code). It’s at the experimental PoC (Proof of Concept) stage. I ran and tested it a lot on my device (8gb RAM, no GPU). On such a weak device, there were no problems.
From me:
If you decide to write a cool skill that solves your problem, please don't forget to share it with me!!!
If you have questions about the project, ask away. I'll be happy to answer!!
bonsuaq•1h ago
I’m sharing this project because I thought such a utility might be useful to someone. I decided to check if local models (SLMs) are capable eniugh for most routine tasks. It is incredibly simple, but the logic of AI Skills is not limited by anything - anything the Python language can do, this product can do. And over time, as local AI becomes even smarter, it will become even more useful.
MacOS already has the Spotlight service. What I created is a kind of analogue but for launching local AI Skills.
Technical part: Python + Tkinter (a bar summoned by the hotkey ctrl + alt) + Ollama (qwen2.5:3b). I specifically didn't use any other heavy libraries. The project turned out very lightweight. One Click install and start .bat file.
Skill Logic: In the root of the project, there is a /skills folder. Each .py file is a script. 1 .py file = 1 AI Skill (Command, Script, etc.). This means that skills can be shared. Like Mods in Minecraft.
Why am I convinced that someone needs this? News recently came out about skills inside the Chrome browser. And also OpenAI's purchase of Sky for Codex. All this news is somehow related to skills and agents. And 100% everyone will eventually need an absolutely free analogue. One with which you can process even thousands of lines of text without paying a penny.
How does it work? Copy the text you need to work with. Press the hotkey (ctrl + alt). And select the desired script (command). The script pulls the text from the clipboard and executes its logic, then returns it directly to the interface without breaking the user's workflow (I attached a GIF demonstration of the project's operation on GitHub). That is, you won't have to use a browser or make other extra movements. All you need is Ollama already installed.
About the Skills I created several very simple skills that I needed most often.
List:
Fix Code - you feed it broken code, you get fixed code
Improve Text - rewrites text cleaner and more professionally
JSON Format - fixes and formats messy JSON
camelCase - text to camelCase
snake_case - text to snake_case
Slugify - text to url-slug
Current project status: Right now the project is very raw and primitive, especially the skills and their logic (literally 10-20 lines of Python code). It’s at the experimental PoC (Proof of Concept) stage. I ran and tested it a lot on my device (8gb RAM, no GPU). On such a weak device, there were no problems.
From me: If you decide to write a cool skill that solves your problem, please don't forget to share it with me!!! If you have questions about the project, ask away. I'll be happy to answer!!
[GitHub Link]: https://github.com/adrianium/Scryptian