AI-Shell Used AI
--- ABOUT ---
ㅤ
An agentic AI terminal. It can automate / do everything, that you could do with a terminal.
(Demo video on GitHub!)
ㅤ
--- EXAMPLE ---
ㅤ
> Create a new Git repository here and add a standard .gitignore file for a godot 4 project. Also, create the skeleton of the project with folders for everything that a 2d multiplayer rouge-like needs.
ㅤ
The AI will then first ask what the project name should be, then use the `git` command to create the github repo, and create the .gitignore file, then use `mkdir` to create all needed folders. When finished, it will provide a complete summary of everything that it has done.
ㅤ
You don't need to worry about any unwanted commands, as it will ask for permission before executing any command. If you find it wrong, you can decline it, and tell it why you declined it to guide it onto the right path.
ㅤ
-- FEATURES ---
ㅤ
- Autonomous command execution step by step
- Web search for gathering recent information (through Tavily)
- Conversation management, and auto saving of current conversation upon interruption
- Incognito mode, that runs local model, and doesn't save conversations (for something like ethical hacking, which most models would refuse to do)
- Compatible with any openAI API compatible model and provider (OpenRouter recommended)
- Both AI requests and direct command execution (prefix anything with `!` to run it as a command)
- A /compact command which shortenes all previous command outputs, to save tokens (money) if there are unneccessarily long text chunks.
- And more! (full list on github)
Followers
Eli Nelson
Check their projects out: Turtle Casino, FishByte, DriftCLI, EmpireForge Discord Bot, PyEvolve, Git Pull News, Rubber Mod, Once-Upon-a-D-j-Vu, Personal Project
Ship Your Project
Get ready!
Once you ship this you can't edit the description of the project, but you'll be able to add more devlogs and re-ship it as you add new features!
Timeline
While the current shipped version is waiting to be certified, i have continued to work on the new one, there is no flashy new stuff (yet!), but im making progress.
I have started to rewrite most of the code, it got really messy, this will probably take a ton of time, but i will try to ship it first with the current code and later reship it with the new code.
Added an 'Incognito' mode, use it for whatever you want ;)
Also refactored the whole codebase basically. There are basically no bugs (that i know of) right now.
I have added better interactive command support, and support for colored command output. it should be working as a complete terminal from now on.
How will you prevent the AI from writing damaging commands? (i.e. “rm -rf” or anything that deletes unwanted directories)
Added much better recent conversations, and better web searching results.
First release version is very close!
I have been working on it for around a month, but i just now stubled up on hackclub. I will document everything here from now on, it is really close to the first complete version.
It supports all openai compatible models, which means it should work with ollama, and yes, it is linux only at the moment, i am releasing the first version in a few days at most.
I just looked closer, is this for Linux?
Also does it support other API than gemini?
Local Ollama model support would be amazing!
Where is the repo? I would LOVE to check this out! Does it support Ollama self hosted models? That would be amazing API keys would be cool too!
There is a confirmation dialog before running any command like:
Execute? [Y/n]
So that if the ai tries to run any damaging command, you can just cancel it. It also asks the user for the reason of decline, which means you can correct the ai to do the right thing.