Malama

Native Linux chat client for local LLMs — no cloud, no browser, no compromise.

---

🚀 Key Features

• Ultra-Minimalist Footprint: Idles at ~20MB of RAM. Bypasses bloated Electron wrappers, ensuring your hardware resources are strictly dedicated to model inference.
• Persistent Sessions (New in v0.2.0): Native sidebar management. Pin, rename, delete, and instantly switch between historical chats without memory spikes.
• Native Markdown & Syntax Highlighting (New in v0.2.0): Custom, zero-dependency Markdown pipeline featuring a dynamic, JSON-pluggable syntax registry (Semantic highlighting for C++, Rust, Python, JavaScript, CSS, and more).
• Low-Latency Streaming: Asynchronous, non-blocking TCP socket implementation for real-time code and text generation.
• Privacy-Focused: A local-first architecture—your data never leaves your machine.

---

🎥 Introduction

See malama v0.2.0 in action:

[Video of using_malamav0.2.0]

Your browser does not support the video tag.

Watch the malama v0.2.0 introduction

🏗 Architecture & Engineering

malama is engineered for stability and speed. Key architectural pillars include:

• Zero-Overhead Storage: Employs an append-only JSON Lines (.jsonl) architecture. Paired with glaze for compile-time JSON reflection, reading and writing massive chat histories operates at O(1) memory allocation.
• Dynamic Grammar Engine: A custom state-machine driven Markdown parser that isolates syntax rules from the UI, allowing pluggable JSON grammar overrides without C++ recompilation.
• Memory Safety: Strict std::nothrow allocation strategy and smart pointer lifecycle management ensure the application remains stable under extreme system memory pressure.
• Concurrency: Uses dedicated background std::jthread workers and a Boost.Asio event loop to manage high-throughput socket communication and disk I/O without dropping UI frames.
• Event-Driven UI: Implements a decoupled, thread-safe global event bus (wxQueueEvent) for cross-thread GUI updates, completely avoiding race conditions.

---

🛠 Build Requirements

malama is designed exclusively for native Linux environments. You will need:

• Compiler: A C++23 compliant compiler (GCC 13+ or Clang 16+).
• Dependencies:
  • wxWidgets (3.2+ or 3.3)
  • Boost.Asio (Network/Concurrency)
  • Boost.UUID (Session ID Generation)
  • spdlog (Logging)
  • glaze (Compile-time JSON serialization)
• Build System: CMake (3.28+)

---

⚙️ Build Instructions

# Clone the repository
# Clone the repository
git clone https://github.com/Magpiny/malama.git
cd malama

# Configure and build
mkdir build && cd build
cmake ..
make -j$(nproc)

# Execute
./malama