Ipnops runs simulations, training, predictions and intelligence, and the interfaces around them.
Ipnops provides compute and interfaces for robotics, machines, infrastructure and software. It integrates with Living Worlds for simulation, training and lives in digital worlds. A job names a kind, runs on Ipnops GPUs or CPUs, takes typed inputs and returns typed, signed outputs.
Worlds to test in before the real thing.
Simulation jobs take geometry, materials, weather and configuration, and return fields, series, renders and scores.
Room impulse responses and the standard room-acoustics parameters per octave band (reverberation time, decay time, clarity, definition) for studios and venues, from geometry and materials.
Temperature, comfort, daylight and air quality for a described room, from its walls, openings and weather.
Agents trained and rolled out among simulated crowds, and body pose read from images.
Forecasts from station and gridded data, with prediction intervals.
Rigid, cloth and deformable bodies, robot-learning tasks with policy export to ONNX, and surrogate models for slow physics.
Scenes rendered, reconstructed and generated from a description or a capture, and a gravity-consistency check on simulated or generated motion.
Make a model fit the job.
Training runs on Ipnops GPUs. A client dataset stays on Ipnops servers.
Low-rank (LoRA) fine-tuning of language models on a client dataset, scored on a held-out set.
A fast stand-in for a slow simulation or a tabular process, fitted and scored on held-out cases.
A forecast with its interval.
Series in, forecast and interval out.
Time series from sensors, meters and logs, forecast with prediction intervals.
Learned approximations of a physical process, fitted on simulation output and scored on held-out cases.
Language, sight and sound, served next to the simulations.
Models run on Ipnops servers inside isolated environments.
Chat, tool use and agents, run in isolated environments under a signed policy.
Embeddings and re-ranking for search over a client's own documents.
Detection, tracking and body pose from images and video.
Transcription, speech synthesis, loudness measurement and speech-quality scoring.
How a measurement gets in and a result gets out.
Sensor nodes sign their readings. Anything that moves a machine goes out through a gate and a person.
Sensor nodes and small computers sign their readings with a device key and send them over Bluetooth LE, Wi-Fi or wired buses.
Each reading carries a grade for how its source holds its key, and a result takes the grade of its weakest input.
Readings and forecasts in and out through the open SensorThings model.
OpenUSD, glTF, MCAP, rosbag, ONNX, Parquet and the common 3D and audio formats.
A REST API, an MCP server and an open SDK client under Apache-2.0.
A job names a kind and gets a typed result. No code travels with it.
Simulation, life-course and scoring jobs for Living Worlds run through the same interface, with results returned signed.
Tell us the job.
Name the area and what you want at the end. We scope the inputs, the run and the output with you.