Built with MIND

Real-world applications built with MIND — from GPU computing to deterministic AI.

Mind-Ray

GPU Computing

GPU Path Tracer

High-performance path tracer with BVH-accelerated CUDA backend. Benchmarked against Mitsuba 3, Cycles, Falcor, and LuxCore.

Performance10-50x on recorded benchmark scenes vs reference engines48x steady-state speedup vs Mitsuba 3 (recorded hardware)

View on GitHub

Game AI

NNUE Chess Engine

GPU-accelerated chess engine with Stockfish-compatible NNUE evaluation, Syzygy tablebase support, and advanced time management.

Features+600 Elo with NNUE evaluationGPU-accelerated search via MIND Runtime

Play Live GitHub

Fractal Voyager

WebGPU / WebGL2

Real-time Fractal Explorer

Explore infinite mathematical complexity with GPU-accelerated rendering. Dive into Mandelbrot, Julia, and Burning Ship fractals with buttery-smooth zoom and pan controls.

FeaturesWebGPU compute with WebGL2 fallbackAudio-reactive colors from microphone input

View on GitHub Live Demo

Substrate Stitch

WebGPU + CPUWebNN / NPUByte-identicalQ16.16

Cross-Substrate Byte-Identity

One deterministic MIND integer kernel, run on two different processors — your GPU (WebGPU) and your CPU — inside a single browser tab, then stitched into one image. Bit-identical output means a seamless image, a pitch-black XOR-diff, and an intact SHA-256 seal. Flip a single bit in one path and the image tears, the diff erupts, and the seal snaps red. A float kernel could not do this. It also runs a hybrid WebNN pass on your NPU/GPU where the browser exposes one — shown honestly, never in the seal.

The wedge, made visibleGPU ⟷ CPU: zero differing pixels, verifiable offlineHybrid WebNN / NPU run where available — honest, never sealedReal crypto.subtle SHA-256 over the raw output bytes

Live Demo

Swarm Brain

WebGPUWebNN

Cognitive Runtime Demo

Interactive WebGPU particle simulation demonstrating MIND's cognitive cycle. 5-invariant verification runs every frame, saga rollback restores state on violation, and a SHA-256 hash chain provides tamper-evident audit. Seeded replay for reproducibility.

ArchitectureSense → Think → Act → Verify → Learn cycleGPU compute + verify pipeline, optional NPU acceleration

Live Demo

Mind-Mem

AI Memory

Memory for AI Coding Agents

Drop-in persistent memory shared across all AI coding clients. One workspace powers Claude Code, Codex, Gemini, Cursor, Windsurf, Zed, and OpenClaw simultaneously.

Powered by MINDOptional MIND scoring kernels — historical per-kernel speedups are recorded in the mind-mem README benchmark table, not a current end-to-end receiptZero core runtime dependencies · PyPI 5.0.1 (source 5.0.2 at 698310d, unreleased)

View on GitHub

Mind-Nerve

Intent Router

Intent-Classification Preselector for Agents

Open the library, hide the cost. A small, fast classifier sits between the user request and the runtime, returns the top-K relevant skills/tools/agents, so the downstream LLM never sees the full library in its system prompt. Library size decouples from token cost.

0.3.1 on PyPI (current release)Phase-1 v1.1-oss evaluation: 96.06% top-5 on 11,922 routing candidates (1,193 held-out eval pairs; 2026-05-16; evaluation manifest; model source)Native Q16.16 encoder ships in the Linux wheel; end-to-end latency is not published and the open-source release has no GPU backendPer-prompt hook routes a 1,300+ skill hub for ~2k tokens instead of a ~95k-token announce · 20-CLI installer matrix with verifymind-nerve acquire: discover external skills/MCPs, quarantine + fail-closed malware vetting, hash-manifested install, live reindex · full .mind tree compiles + executes on mindc 0.10.2pip install mind-nerve · weights auto-download · Apache-2.0

GitHub PyPI HF

Mind-Inference

LLM Inference

MIND-Native Inference Pipeline

Paged KV cache, FlashAttention-2, speculative decoding, continuous batching, disaggregated prefill/decode, and 512 governance — compiled to native ELF. Deterministic by default (CALF 1.0).

Compiled MIND30 TFLOPS matmul · 3.48M TPS with governance12 modules · 5/5 tests · 512 kernel inline

Live Demo

GEMM Bench

WebGPU Benchmark

In-Browser GEMM Benchmark

Apples-to-apples WebGPU matrix multiplication benchmark. Compares MindLang AOT-compiled WGSL shaders against ONNX Runtime Web on your GPU, in your browser.

Performance7-19x faster than ONNX Runtime Web~4.5 TFLOPS at 4096x4096 on WebGPU

Run Benchmark Source

Arch-Mind

Architectural Governance

Deterministic Code Scanner

Nine Q16.16 fixed-point metric kernels over a typed dependency graph. Deterministic Python reference scorer, cryptographically-chained session deltas, rules declared in .mind — not CI warnings. The time-travel scoreboard reads a dated, checked-in fixture scan of star-ga/mind-mem (not a live feed).

STARGA Commercial9 metric kernels · 4 tree-sitter languages + heuristic MIND · HMAC-SHA256/Ed25519 evidence logBit-identical reproducibility across machines

Live Demo

512 War Room

Governance

Adversarial Governance Demo

Can you break legitimacy? Launch 6 adversarial attacks against a live DIFC governance system — forge constraints, suppress signals, replay evidence, flood the pipeline, collude. Watch every attack fail in real-time, with cryptographic proof anchored to Arbitrum.

Pure MIND44 modules · 6 attack vectors · L2 anchoredStory Mode + Free Attack + Regulatory Lens

Live Demo

Building something cool?

We'd love to hear about your project. Submit your story to be featured in our showcase and newsletter.

Submit Your Project