Open a CSV, a sensor on a serial port or a camera. Chart it, track people and rooms with models that run on your own machine, and script every run in Python with no window. Then put the result on a screen, a projector or a wall of lights.
Every picture on this page is the app's own output.
The jobs that come up week to week, and how each one is done here. Every step stays a node you can open again when the data or the question changes.
Open a CSV or JSON file, compressed CSV included, and let it reload on its own when a new export lands. Filter, sort and pivot a table, and roll groups up to sums, means, medians, standard deviations and percentiles. The viewport shows any table or channel as a table, a graph, a histogram or a summary.
Plot channels as lines, stacked lanes, a scatter with its correlation, a histogram, a heatmap or a spectrum, up to 8192 pixels a side. Draw bar, pie and scatter charts as SVG. Save a still as PNG or TIFF with its print size in dpi, as float EXR, or as a numbered sequence for a video.
Find 17 body points per person, or 133 with the face, feet and hands, from an ordinary camera or a video file. Track moving blobs with steady IDs, draw zones to count who enters, leaves and stays, and build a heat map of where people went. The models run on your machine, so footage never leaves it.
Read color, depth and IR from a depth camera, or work out depth from one webcam frame with a local model. Turn either into points, walk a camera through a room to map it, and measure area, perimeter and volume.
Play recorded interviews or a talk into a speech model on your own machine: a small one on a laptop, a large one on a graphics card. Pick the language or let it listen for one. It waits for a pause, so sentences are not cut in half.
Run a graph once per file or once per value, every combination, each run its own process and as many at once as you allow. On the virtual clock a run steps itself one frame at a time, so the same graph does the same work at the same timestamps.
$ kinematic run lorenz_runs.kin \
--each data.file_path='runs/*.csv' \
--set out.file_path='out/{stem}.png' --jobs 3Sound lives in the same file as the data. Play a series as sound, let a channel steer a synth, or go the other way and read a recording as numbers you can chart and save.
The graph keeps running, so the file that charts the data can also put it in front of people: on a projector at a talk, across a wall of lights at an open day, or around a room in an exhibit.
Build the system you are explaining and watch it run: water in a tank, bodies under gravity, a pattern growing out of two chemicals. Sweep any setting from the command line to cover the whole range.
Bring in a Gaussian splat or a point cloud of a site, a find or a specimen. Clean it up, check what it is made of, measure it and render it, or turn it into a solid mesh for other tools.
Tables, sensors, cameras and results come in and go out in the formats your lab already uses, and the whole thing can be driven from code.
A node can be a Python class with your own ports, and a full API outside the graph opens files, changes settings and reads results. This one loads a numpy file and hands the graph a table.
import numpy as np
import kinematic
class LoadTrial(kinematic.node.PythonScriptNode):
type_name = "LoadTrial"
memo = "source" # runs once on its own when the graph starts
outputs = [kinematic.node.PortDecl.typed_data("Table_Out")]
def execute(self, ctx):
a = np.load("trial_07.npy") # t, x, y, z for each sample
table = {c: a[:, i].tolist() for i, c in enumerate("txyz")}
ctx.post_output(0, kinematic.values.Value.typed_data(table))kinematic pip installadds the packages you already use to the app's own Pythonkinematic repl --grapha Python prompt with no window, bound to a graph you can start, change and readkinematic py-testruns a Python test suite against your graphsThe file you build in the app runs from a terminal, over SSH or on a lab machine with no screen. Sweep a setting, run once per file, and keep every value that came out.
$ kinematic run trial.kin --sweep noise.seed=1,2,3 --jobs 3
$ kinematic run trial.kin --clock virtual --frames 600
$ kinematic run study.kin --relay-dir out/ --until-dead
$ kinematic test study.kin --expect-output "done"--sweeponce per value, every combination, each run its own process--clock virtualthe same work at the same timestamps, repeatable on the same graphics card and driver--relay-direvery output into a folder: pictures as files, channels and tables as NDJSON--offlinesealed off the network: no model downloads, no pip. Models you already have still runtestchecks what came out, with a JSON report and an exit code for CIShort lessons that each end with something finished, and demos that come with the app. Take one apart and keep the parts you need.
Develop · 12 minBuild a ring of spheres with a Python scriptTwelve spheres around a wavy ring, every one placed by a short script that adds the nodes and wires them.
Develop · 8 minSpin a donut with one line of PythonA donut that tumbles and breathes, moved by two short Python expressions typed into its settings.
Develop · 10 minRender a demo to pictures from the terminalEvery frame of a demo rendered to numbered pictures from one command, with no window.
Perform · 12 minMake a visual that pulses to your musicA field of glowing bokeh that zooms in on every loud beat of a song: sound driving a picture.
Simulate · 8 minBreak a dam, then turn the water to mudA column of water lets go, slams the far wall and sloshes back. Then make it thick and slow.
Play · 8 minClean up a field recordingSteady background noise turned down, the rumble gone, the bumps trimmed and a fade at the end.
Triple pendulumThree pendulums swinging from steel anchors, their arcs trailing under a spot light.In the Demos browser
Orbital gravityA swarm of boxes orbiting a center of gravity, tumbling as they fade in and out.In the Demos browser
Wave interferenceTwo wave fields laid over each other into interference patterns.In the Demos browser
Reaction-diffusion artA Gray-Scott pattern moving between spots, stripes and labyrinths as its feed and kill rates drift.In the Demos browser
Cellular life simulationA grid of cells living and dying by the rules of Life, softened into glowing colonies.In the Demos browser
Dam breakA column of water races down the tank, slams the far wall and sloshes back. One solver node runs it.In the Demos browser
Webcam, GPUScan a room by walking through itWalk a camera through a room and the room builds itself out of colored points.In the Demos browser
WebcamWhole body skeleton overlayTorso, legs, feet, face and both hands, 133 points in all, with a running count of the people in frame.In the Demos browserFree. No account, no license key, no seat count, so it can go on every machine in the lab. macOS, Windows, Linux and Raspberry Pi.
