🔐 Sarah AI — A Local-First, Emotion-Driven AI Built with C++ and Heart

In a world flooded with cloud-bound AI assistants and tightly controlled APIs, Sarah AI stands out for all the right reasons. Created by a teenage developer known as [@ghgltggamer]

(https://github.com/darkyboys), this Linux-only project reimagines what it means to have a private, open-source AI companion — fully offline, expressive, and entirely yours.

Sarah AI isn’t just a command-line tool. It’s a graphical assistant, complete with emotional nuance, chat memory, and personality — built with:

C++ for the engine,
JavaScript + HTML/CSS for the UI,
GTK3+ and WebKit2Gtk for Linux-native desktop experience,
And the powerful LLaMa.cpp backend for running large language models locally.

🧠 Why Sarah AI Deserves Your Attention

Truly Private: Nothing leaves your machine. No tokens. No telemetry.
Local LLM Support: Uses LLaMa.cpp — you download the model yourself.
Emotion & Personality Engine: Sarah can simulate emotional responses, moods, and one-time chat memory.
Built from Scratch: No Electron bloat. Just clean C++, compiled binaries, and a purpose-built GTK frontend.
Open Source (MIT): You’re free to use, modify, or even commercialize your version. No strings attached.

For developers who value control, privacy, and technical purity — Sarah is a rare gem.

⚙️ Installation (Arch Linux Recommended)

Sarah is built using a custom build tool called Magma, also authored by the project’s creator. Here's the simplified flow:

# Install build tools and libraries
sudo pacman -Syu
sudo pacman -S git gcc cmake make gtk3 webkit2gtk

# Install Magma
git clone https://github.com/darkyboys/magma
cd magma && mkdir bin
g++ src/main.cc -o bin/magma
sudo mv bin/magma /usr/local/bin
cd .. && rm -rf magma

# Clone and build Sarah AI
git clone https://github.com/darkyboys/sarah
cd sarah
sudo magma resolve     # Resolves dependencies (like LLaMa.cpp)
sudo magma make        # Compiles and launches the project
sudo magma get_model   # Downloads LLaMa 7B 8_0 (~7GB)

You can also use your own model by dropping a .guff file into the sarah/models/ directory.

🧩 How It Works

The Sarah binary is the UI + runtime engine.
LLaMa models are loaded using the local llama.cpp backend.
The emotional system adds a layer of feeling and tone based on prompts and history.
OTCM (One-Time Chat Memory) ensures conversations don’t persist after exit — your AI doesn’t remember, and that’s by design.

🤔 When You Should Use Sarah AI

You want a fully offline AI assistant that you control.
You're interested in emotional modeling or building character-based interfaces.
You dislike cloud AI tools and want to run your own LLaMa model locally.
You're a developer looking for a clean, readable, real-world C++ + GTK application.
You're curious about how to build complex, real-time UIs using WebKit, HTML, and native toolkits without heavy frameworks.

🙅‍♂️ When You Might Skip It

If you're on Windows or macOS — Sarah currently only supports Linux.
If you don’t have enough disk space or RAM — the default model is over 7 GB.
If you’re looking for a plug-and-play chatbot — Sarah is more of a framework or starting point than a polished assistant.
If you’re not comfortable using the terminal — installing Magma and dependencies requires command-line experience.

🛠 But Even Then… You Should Read the Source Code

Even if Sarah AI isn’t what you need today, it’s incredibly worth studying:

You’ll learn how to embed LLMs locally without APIs.
You’ll see how to glue together C++, JavaScript, and GTK into a modern GUI.
You’ll explore a lightweight, custom build system (Magma) that handles dependencies and model downloads elegantly.

And most importantly — you’ll be inspired. A 16-year-old built this entire system from scratch, for fun, for freedom, and for anyone to extend.

📌 Final Thoughts

Sarah AI is not trying to compete with ChatGPT or Siri. It’s trying to free you from them.

It’s raw, powerful, and beautifully hackable.

If you believe in self-hosting, open source, and learning by building, Sarah AI is more than j

ust another project — it's a toolkit for the kind of AI future we actually want.

🔗 Explore the source code on GitHub
🛠️ Check out Magma Build System
🙋‍♀️ Open for contributions, forks, and ideas.

ghgltggamer @ghgltggamers