DEV Community

Javad
Javad

Posted on

Atoqu - The first Atomic-Qunatum Search Engine ever exists in your system!

#atoqu #overlab #programming #opensource

Hey Dev Community! 👋

So… I built something. Something atomic. Something quantum. Something that makes your laptop fan spin like it’s about to take off to Mars. 🌌

Meet Atoqu — the Atomic + Quantum-inspired search engine core in C++.

It’s not just a search engine. It’s THE search engine.

👉 GitHub: https://github.com/OverLab-Group/atoqu

Go star ⭐, fork 🍴, open issues 🐛, submit PRs 🔥, and let’s make history together.

☢️ Atomic? Like Nuclear? But Safer (and Cooler)

“It’s Atomic, like Nuclear, but not! 💣”

No mushroom clouds here — just Atomic LEGOs.

  • Each search mode is an independent atomic unit.
  • You can snap them together like LEGO bricks 🧱.
  • Want keyword search? Plug in LiteralMode.
  • Want embeddings? Snap in VectorMode.
  • Want both? Smash them together in HybridMode.

Atomic = modular, composable, indestructible.

Basically, if LEGO and Iron Man had a baby, it would be Atoqu. 🦾

⚛️ Quantum? Schrödinger’s Search Engine 🐱

Quantum-inspired means parallel evaluation.

Multiple modes exist at once — like Schrödinger’s cat. 🐱⚛️

  • Atoqu evaluates all modes in parallel.
  • Then collapses them into a weighted ranking.
  • Your query decides which mode “lives” and which “dies.”

Quantum = superposition of search strategies.

It’s like having Google, Bing, DuckDuckGo, and your grandma’s recipe book all answer at once. 🍲

🛠️ Core Architecture (v1.2)

  • Engine: AtoquEngine
  • Stores:
    • DocumentStore (tags + metadata)
    • VectorStore (cosine similarity, CPU + optional GPU acceleration)
  • Modes:
    • NormalMode — browser-style, classic search
    • LiteralMode — keyword search
    • VectorMode — embedding-based search
    • HybridMode — combined scoring
    • BM25Mode — term-based ranking
    • RecencyMode — time-aware ranking
    • TagBoostMode — tag-aware ranking
  • Embeddings:
    • HashEmbeddingProvider — deterministic, dependency-free
    • LlmEmbeddingProvider — config-driven, LLM-ready
  • GPU Backends:
    • CUDA — production-ready cosine similarity
    • OpenCL — skeleton, safe CPU fallback
    • Vulkan, Metal, SYCL, HIP — safe stubs, fail-closed

⚡ GPU Acceleration (Turbo Mode)

Here’s the magic: Atoqu can offload heavy vector similarity operations to your GPU.

That means faster searches, less CPU pressure, and more bragging rights. 🏎️

Step 1: Configure GPU backend

Edit config/gpu.json:

{
  "backend": "cuda",
  "dimension": 384,
  "maxdocsper_batch": 8192
}
Enter fullscreen mode Exit fullscreen mode

Supported values:

  • "none" (default, CPU-only)
  • "cuda" (NVIDIA GPUs)
  • "opencl" (cross-vendor)
  • "vulkan" (stub)
  • "metal" (stub)
  • "sycl" (stub)
  • "hip" (stub for AMD)

If initialization fails, Atoqu falls back to CPU-only vector search.

No crashes, no drama. 🎭

🎛️ Modes Configuration (So Simple It’s Almost Funny)

All modes are controlled by config/modes.json.

Here’s the default setup:

[
  { "name": "NormalMode",   "enabled": true,  "weight": 1.0 },
  { "name": "HybridMode",   "enabled": true,  "weight": 1.0 },
  { "name": "LiteralMode",  "enabled": true,  "weight": 0.8 },
  { "name": "VectorMode",   "enabled": true,  "weight": 0.8 },
  { "name": "BM25Mode",     "enabled": true,  "weight": 0.9 },
  { "name": "RecencyMode",  "enabled": false, "weight": 0.5 },
  { "name": "TagBoostMode", "enabled": false, "weight": 0.5 }
]
Enter fullscreen mode Exit fullscreen mode

Want to boost fresh results? Flip "RecencyMode" to true.

Want tags to matter more? Enable "TagBoostMode".

It’s literally editing a JSON file. That’s it. 📝

🧑‍💻 Building Atoqu (All Methods)

Method 1: Classic CMake (CPU-only) 

mkdir -p build
cd build
cmake ..
cmake --build .
./atoqu --http 8080
Enter fullscreen mode Exit fullscreen mode

Method 2: CMake with GPU support 

mkdir -p build
cd build
cmake -DUSE_GPU=ON ..
cmake --build .
./atoqu --http 8080
Enter fullscreen mode Exit fullscreen mode

Make sure your GPU backend is set in config/gpu.json.

Method 3: Docker (CPU-only) 

docker build -t atoqu:cpu .
docker run -p 8080:8080 atoqu:cpu
Enter fullscreen mode Exit fullscreen mode

Method 4: Docker (GPU-enabled with NVIDIA) 

docker build -t atoqu:gpu .
docker run --gpus all -p 8080:8080 atoqu:gpu
Enter fullscreen mode Exit fullscreen mode

Method 5: Docker Compose (CPU + GPU variants) 

docker-compose -f docker-compose.cpu.yml up
docker-compose -f docker-compose.gpu.yml up
Enter fullscreen mode Exit fullscreen mode

Method 6: GitHub Actions / GitLab CI

CI pipelines are already set up.

Push your fork, and tests will run automatically:

  • Unit tests
  • Integration tests
  • GPU-aware tests
  • Sanitizers
  • Static analysis

📚 Documentation

Generate docs with one flag:

mkdir -p build
cd build
cmake -DATOQUBUILDDOCS=ON ..
cmake --build . --target docs
Enter fullscreen mode Exit fullscreen mode

Docs will appear in:

  • Doxygen HTML: docs/doxygen/html/
  • Sphinx HTML: docs/sphinx/_build/html/

🗺️ Roadmap

  • Short-term: finish GPU kernels, expand docs.
  • Mid-term: smarter HybridMode, pluggable storage backends.
  • Long-term: full-featured search service, benchmarks vs existing engines, plugin ecosystem.

Basically: today it’s a search engine core. Tomorrow it’s Skynet. 🤖 (but friendly, promise).

😂 Fun Facts

  • If your laptop fan spins up, that’s Atoqu saying “hi” in CUDA.
  • If it doesn’t, that’s Atoqu saying “hi” in CPU mode.
  • Either way, it’s polite. 😎

🎯 Call to Action (Do It Now!)

Atoqu is open-source, reference-grade, and waiting for your contributions.

  • ⭐ Star it — show support.
  • 🍴 Fork it — experiment with modes.
  • 🛠️ Open issues — ask questions, propose features.
  • 🔥 Submit PRs — tests, docs, kernels, optimizations.
  • 📊 Benchmark it — share your results.

👉 https://github.com/OverLab-Group/atoqu

💡 Final thought:

Atoqu is Atomic (like LEGOs), Quantum (like Schrödinger’s cat), and GPU-accelerated (like a rocket).

It’s more powerful than nuclear bombs — but instead of destruction, it builds the future of search. 💥🧪🚀

Top comments (0)