HAMON Engines, worlds and instruments.

01 · CORE TECHNOLOGY

The HAMON Engine

An AI-native creation engine.

One engine for games and worlds, video and film, animation, sculpting and characters, a full physics engine, and simulation, science and engineering tools, built so AI agents can drive it directly as well as people. It is a workspace of 27 Rust crates with native C++, CUDA, TypeScript and Go components, and it runs natively and in the browser through WebAssembly and WebGPU.

• Built for agents

  One method table served as a command line, JSON-RPC and MCP, so AI agents and tools can build, inspect, step and test the engine with no window. Typed deterministic runtime commands, replay and SHA-256 experiment receipts make what an agent did checkable. The native editor is an optional client.

• Deterministic core

  An archetype ECS with a work-stealing job system, a fixed timestep and seeded randomness, so a run replays exactly from its recorded input. Lockstep input-frame networking builds on the same replay model.

• A full physics engine

  Rigid bodies with hinges, motors, capsules and triangle surfaces; elastic strands and soft membranes (XPBD, with tearing and plastic yield); deformable terrain driven by living state; and analytical engineering solvers for hydraulics, cross-sections and rods. Native C++ solver adapters cover a MuJoCo bridge, foot-lock IK, reachability analysis and system identification.

• Animation and characters

  Skeletal animation with blending, IK, retargeting, morph targets, squash and stretch, and follow-through. Code-only sculpting from signed-distance shapes builds characters, creatures, birds, insects, trees, flowers and whole scenes, with no external modelling tools.

• Video and perception

  Native C++ modules for media ingest, camera calibration, ONNX-based tracking and a world compiler, plus CUDA kernels for skinning and splat viewing. They power the video tools below.

• Rendering and style

  A real-time forward PBR renderer with cascaded sun shadows, point lights and height fog, headless and hardware-qualified capture, and a style system with 17 selectable looks, from hand-painted luminous cel with ink line art to watercolor and graphic ink. Extruded 3D text from font outlines is built in.

• Assets, materials, audio and scripting

  Diffable text scenes and memory-mapped cooked packs, glTF import and an asset journal; half-edge mesh authoring; node-graph materials that compile to WGSL; a deterministic audio mixer; sandboxed WebAssembly scripting; and HamonLang, an experimental systems language.

• Science and engineering tools

  Content-addressed experiment history in Rust, TypeScript and Go, a Studio UI, an Innovation Lab integration, and runnable studies in surface diffusion, radiative cooling and electrical networks.

In progress. The AI-native authoring roadmap is open work: live authoring connected to rendering, material graphs executed in the scene renderer, procedural modeling that updates only the affected geometry, a high-quality final-image workflow, and discovery of tools and reusable assets.

The engine source is private; its work is published through the games and tools on this page.

03 · VIDEO, FILM AND MOTION

Tools for working with moving images

Creation and analysis tools for video, from new camera angles on ordinary footage to forecasts replayed against what actually happened. They are early and mostly private; each says plainly what it does not claim.

• Phantom Window

  Walk a virtual camera through ordinary video. It estimates depth per frame, recovers the camera path, and renders new viewpoints on the GPU, marking which pixels were observed and which inferred.

• Replay Forge

  A basketball replay studio: connect a discussion to the relevant possession, branch the replay, and keep every claim bound to the source video.

• Captionverse

  Spatial, speaker-aware captions and sound cues for video, with hash-matched cue manifests and local speech recognition that produces reviewable drafts.

• Chronos Vision

  A temporal observatory for video: color-track measurements, three contrasting future-trajectory hypotheses, and past forecasts replayed against later frames. An experimental first release.

• Mechanism

  A motion lab for machinery. Track markers in real video, compare image-plane paths and fit constrained kinematic models recomputed from saved evidence.

04 · SCIENCE AND RESEARCH INSTRUMENTS

Instruments for looking closely

Evidence-first tools that share one approach: show what was observed, keep the uncertainty, and refuse to overclaim. The repositories are private and have no public demo yet.

• WorldLaw

  Learns executable physical laws from observations and actively tests the resulting hypotheses.

• Afterimage

  A physical-world evidence engine. It reconstructs what is known and what might be true from fragmentary media, keeps competing explanations, and says what to observe next.

• Synapse

  A spatial observatory for real running computational systems. The first slice is a local, read-only browser instrument for a live process tree.

• Microverse

  An early investigator for microscopic life: follow individuals through supplied detections and keep competing explanations of an encounter.

• MATERIALTRUTH

  A photometric inspection desk for surfaces. It infers relative albedo and normals from multi-illumination images and refuses where the data cannot support a result.

05 · AGENTS AND WORKSHOP

The rest of the estate

The tooling that lets one person run all of this. Both are private and early.

• HAMON agent platform

  Early work on a supervised agent runtime: durable claims, recovery and replay, and memory with explicit rights. Built as deterministic fixtures; not a deployed product.

• HAMON Workshop

  A simulated workshop operated through MCP: discover, reserve and run a virtual workcell, keep failed actions, and recover from the last committed pose. A software foundation for physical capability, not hardware yet.