Products

A from-scratch game engine built with C++20 and Vulkan. Features a deferred PBR rendering pipeline, ECS architecture with explicit system lifecycles, a particle VFX system, a full asset pipeline with custom importers and VPak packaging, and a deterministic job scheduler. Used to power the C.A.T. game and available for source licensing or consulting engagements.

The asset pipeline for the V Engine. VPak Builder creates parallel zstd-compressed archives. Asset Importer converts GLB/glTF to engine-native Vmesh/Vtex, WAV to compressed VPak entries, images with mipmaps, and safetensors to .vnn/.vvo/.vemb. Asset Viewer provides an interactive Vulkan preview with PBR materials and animation playback.

A standalone editor for authoring and previewing particle effect presets. Features a live Vulkan-rendered viewport, editable property panel, preset management, playback controls, and ribbon/trail systems. Outputs TOML presets consumed by the engine's VFX system.

A skeletal animation authoring and pose manipulation tool with viewport preview, timeline editor, pose manipulation with IK, skeletal hierarchy navigation, and animation blending. Supports the engine's animation runtime and asset pipeline.

An interactive CLI and training harness for a GRU-based continuous-output chat model. Features BPE tokenization, dual-stream attention decoder (MiniLM sentence embedding + context buffer), GPU compute support, and configurable hyperparameters. Suitable for on-device NPC dialogue and agent reasoning.

A personality catalog generator producing MBTI-based character profiles for AI agent characterization. Combines Plutchik emotion weights, MBTI types, cognitive biases, and Dale Carnegie principles into exportable persona profiles for the dialogue and NPC systems.

A retainer-based AI safety service for organizations running autonomous agents, hybrid human-AI systems, or networks where AI has operational authority. When an AI goes rogue — hallucinating at scale, abusing tools, escalating privileges, or running amok across your sub-nets — Guardian Angel deploys containment, not checks. We service agents and humans alike, across any hybrid system, machine, or operational surface. Standard retainer: $10,000. Deployment billed at $1 per token. Includes one free consult hour per month and one free hour of initial syndrome diagnosis and health checkup for any hybrid system, machine, or human in your fleet.

A security service offering based on the C.A.T. crypt, port-guard, net-ctl, and pentest tooling. Includes a monthly Game Studio Security Threat Brief PDF, one-off security reviews of build/pipeline/servers, and ongoing hardening retainers.

High-end consulting for studios stuck on Vulkan, ECS architecture, GPU performance, or RL training pipelines. Offers 1-2 week optimization sprints, GPU RL training setup (PPO/SAC/DQN, decision transformer), NPC behavior architecture reviews, and formal-methods reviews linked to the Lean 4 proofs.

C.A.T. has published research papers, Lean 4 formal proofs, and benchmark datasets suitable for non-dilutive grants and partnerships. Targets include Epic MegaGrants, NSF SBIR/STTR, DARPA/IARPA, UK Games Fund, Canada Media Fund, and Nvidia/AMD/Intel compute grants.

The primary game being built on the V Engine. Currently at M10 with a Steam Next Fest demo targeted for October 2026. The demo requires the Must Level chain, the Naxzyu encounter, and a trailer. Not a near-term revenue source, but the long-term upside of the project.

A neural network inference benchmarking and edge deployment tool. Provides per-stage timing (tokenization, embedding, QKV GEMM, attention, FFN, pooling), telemetry CSV output, accuracy verification, configurable iterations, and memory tracking.

A suite of authoring tools including the Level Editor (GUI viewport, entity browser, property panel, collider editing, and level packing CLI), the Tileset Editor (WYSIWYG tileset definitions with TOML export), and the Dialogue Editor (visual node-based graph editor for branching dialogue trees).

CIC turns verbose Markdown instruction files into compact, verifiable programs. It parses .md into a line-oriented PIR intermediate, compiles back to compact .md, and runs security, lint, and semantic regression checks before install. The pipeline detects prompt injection, stealth directives, exfiltration, leaked secrets, XSS, malware, untrusted external instructions, and over-privileged commands mapped to the OWASP Agentic Skills Top 10. It validates frontmatter, heading structure, naming conventions, and CommonMark syntax across skills, workflows, abilities, rules, memories, plans, and AGENTS.md. A native `cic` executable and Linux system command are scheduled for release by end of week.