No physical site. No complete CAD. High consequence decisions.
Teams must evaluate rover paths, crane placement, solar arrays, and robotic assembly before hardware reaches the surface.
Hylos Labs
Semantic spatialintelligencefor agents.
Hylos compiles real and generated worlds into structured 3D object spaces agents can reason over, edit, simulate, and control.
Compilation Layer
Compiling visual representations into spatial programs agents can
Hylos sits between an agent's tools and its visual representation of the world, turning scenes and generated concepts into semantic-spatial programs: objects, parts, surfaces, topology, constraints, and relationships.
Agents do not just observe space through Hylos. They can generate new structures, modify existing environments, and operate on parts, surfaces, and relationships with addressable control.
Interactive Demo
Watch the compilation layer become an action.
A visual representation is converted into addressable spatial structure. Once the layer is ready, the agent can target the cabinet and execute the interaction.
Agent rationale
Waiting for semantic layerUser prompt
Open the highlighted cabinet door.
Current state
Building semantic object space
Reasoning over scene graph
Looking at reference video
Generating topology
Creating object space
Compiling action target
Executing cabinet interaction
semantic_layer
Semantic layer0%
Agent action0%
Concept Case Study
How WordLabs and Hylos simulate a lunar build site before it exists.
WordLabs generates a high-fidelity environment from mission constraints. Hylos compiles that world into structured topology, where every surface, object, and constraint is addressable by an agent.
View the full case study
WordLabs generates the world. Hylos makes it operable by agents.
WordLabs produces the environment. Hylos compiles it into an agentic spatial model, extracting parts, surfaces, topology, constraints, and relationships so agents can reason about and modify the scene.
Spatial planning becomes an executable system.
Engineers do not just review scenes. They instruct agents to test layouts, validate constraints, and iterate designs before deployment.
Agent Operation Layer
Generate a worksite. Let agents operate on it.
Not just a model. A space agents can operate on.A planner defines constraints: access, coverage, shielding, and clearance. WordLabs generates the environment. Hylos compiles it into an agent-ready spatial model where every object, surface, zone, and constraint is addressable and modifiable.
Create high-fidelity mission environments before any physical capture exists.
Extract parts, surfaces, topology, constraints, and relationships. Make everything addressable.
Agents plan, edit, and validate layouts across terrain, modules, paths, and constraints.
Export validated configurations to CAD, simulation, and robotics systems.
Applications
Agents need spaces they can generate, edit, and operate on.
Manufacturing
Generate, inspect, and reconfigure production cells, fixtures, parts, and robotic workspaces as agent-operable spatial systems.
Learn more →Defense
Compile terrain, facilities, assets, and mission concepts into structured environments for planning, simulation, and agent coordination.
Learn more →CAD / Design Systems
Turn prompts, sketches, scans, and references into editable 3D object spaces with topology, constraints, and part-level structure.
Learn more →Case Studies
From scene understanding to spatial action.
Scene Understanding
01Manipulable environments
Convert raw scenes into object-aware, surface-aware, constraint-aware spaces that agents can work inside.
Spatial Action
02Agentic control
Let agents reason about where things are, what they mean, and how they can be changed.
Technology
A semantic-spatial layer for agentic systems.
Hylos converts visual inputs and generative outputs into semantic-spatial structure: topology, surfaces, parts, object relationships, and executable spatial operations.
Early system. Built toward agentic spatial control.
- Generative 3D object spaces
- Semantic topology and part structure
- Agent-directed spatial operations
captureunderstandplanact
Build agents that can work with the world.
We're developing the spatial layer that lets agents reason about, operate on, and eventually manipulate real environments.
Contact us