Pluton is an independent systems programming language created by Artur. Featuring native vector math, a 60 FPS 2D/3D engine, low-level binary networking, and an active initiative training a foundation language model directly in Pluton — accelerated by Google Cloud & Frontier Google AI.
# Native 2D vectors and direct hardware access
val author = "Artur"
val start_pos = (10, 20)
val velocity = (5, -2)
val new_pos = start_pos + velocity
when (hardware.is_available()) {
val dev = hardware.gpu(0)
echo("GPU detected:", dev.name)
echo("VRAM available:", dev.vram, "MB")
} otherwise {
echo("No GPU detected.")
}No mockups or artificial benchmarks. Inspect the exact code files written in Pluton, or toggle the AST Inspector to verify the parsed syntax tree generated by the compiler front-end.
We are currently engineering a domain-specific large language model designed and executed natively within the Pluton language runtime. By discarding the Python Global Interpreter Lock (GIL) and runtime interpreter layer, Pluton-LLM operates directly against GPU memory allocations via hardware.gpu() with deterministic execution cycles.
Traditional PyTorch/Python loops suffer from CPU launch overhead. Pluton's tight loops execute tensor operations with zero interpreter garbage collection interruptions.
We leverage Frontier Google AI via Google AI Studio / Vertex AI to generate high-fidelity synthetic reasoning corpora, mathematical logic puzzles, and domain-tailored training pairs.
Google Cloud TPU / GPU infrastructure acts as our compute backbone — validating gradient backpropagation steps, detecting loss divergence, and proving convergence invariants.
Training an LLM on an independent language stack requires vast compute and reasoning throughput: validating custom tensor autodiff kernels, synthesizing billions of tokens of specialized syntactic training data, and real-time loss stabilization. Google Cloud compute infrastructure (Cloud GPUs / TPUs) and Frontier Google AI's massive context window provide the deep architectural support indispensable for this level of systems engineering.
Consistent, predictable CLI tools built for speed and clarity.
pluton run <script.pluto>Launches the Pluton runtime environment, initializes requested hardware or 2D/3D windows, and executes the file at 60 FPS.
pluton check <file.pluto>Runs the front-end parser and type validator without executing side-effects. Reports syntax errors, unbound symbols and vector mismatches.
pluton train --model <cfg>Triggers the native neural training pipeline with GPU memory telemetry and automated loss logging.
Every library below is fully demonstrated in our real working code samples.
Windowed 2D/3D graphics with a hardware frame limiter. 2D: draw_rect, draw_round_rect, draw_text, draw_grid, draw_glow_circle, draw_gradient, beep. 3D: set_camera, set_light, draw_cube, draw_pyramid, draw_grid_3d.
Low-level TCP client with a ByteStream for writing integers, longs and strings, building custom binary packets and parsing packet headers for game protocols.
Query whether dedicated accelerators are available with hardware.is_available() and inspect their name and vram directly through hardware.gpu(i).