What Minimalist Physics Actually Is
It is a lightweight physics sandbox environment that runs in your browser or as a standalone desktop app. The core concept strips away everything except the essential mechanics: rigid bodies, springs, friction, gravity, and collision resolution. There are no storylines, no progression systems, no point-scoring. You place objects and let the engine do its thing. I spent about three weeks last year really digging into this because I needed a simple tool for a visual project that required realistic but unpolished physics interactions. Most people coming from more established sims like Algodoo or Box2D playgrounds find the interface jarring at first. It deliberately refuses to hold your hand.
Minimalist Physics Free Download
The official source for the desktop build is on their GitHub repository. The web version works without installing anything, but the desktop release is worth getting because it handles larger scenes without the memory throttling that browsers impose. Grab the latest stable build from the releases page and run the installer. On Windows it drops a portable .exe that you can move around. On macOS it is a standard .dmg. Linux users get an AppImage and a flatpak option. The install process takes about forty seconds. No account required. No telemetry you have to opt out of. That alone is unusual these days.
How It Actually Works Under the Hood
The engine is built on a modified Verlet integration scheme rather than the more common impulse-based rigid body approach. This means constraints are handled implicitly through position corrections rather than force applications each frame. The practical result is that stacks of objects tend to settle more naturally without the jittery micro-bouncing you see in impulse solvers. It also means the simulation is slightly less accurate at high velocities, which matters if you are simulating projectile impacts or fast-moving colliders. Another thing most users miss: the friction model is Coulomb-based with a separate static and kinetic coefficient that you can tune per object material. The default values are reasonable for generic scenarios but will look wrong if you are simulating something like ice or rubber. I ran into this early on when I tried to recreate a sliding puck scenario and the objects stopped dead way too quickly. The fix was setting the material friction coefficient to 0.02 on the puck and 0.8 on the surface it was sliding across. Without that separation everything feels like it is moving through wet concrete.
Get the Full Details
The Real Limitations
This is not a substitute for an engineering-grade simulator. The collision detection uses broad-phase sweeping with narrow-phase SAT (separating axis theorem) for convex shapes only. That means you cannot create concave meshes without manually decomposing them into convex hulls. If your workflow involves complex organic geometry, you will hit a wall fast. For simple boxes, cylinders, and wedges it is perfectly adequate, but anything requiring triangulated meshes will frustrate you. The solver also does not support continuous collision detection. Fast-moving objects can tunnel through thin walls if their velocity exceeds the threshold where the discrete steps skip over the geometry. I discovered this when testing a high-speed pendulum experiment where the arm was passing through the floor geometry at certain angular velocities. The workaround is reducing the time step or increasing the iteration count in the solver settings, but both choices come at a performance cost. A scene that ran smooth at sixty frames per second on default settings dropped to twenty-four when I cranked the iterations up to resolve the tunneling issue properly. There is also no built-in scripting language. You can define initial conditions, set material properties, and adjust parameters through the UI, but you cannot write custom behaviors or procedural object generation. If you need parametric experiments you either export the configuration files manually between runs or write an external script that manipulates the saved state files. The file format is JSON and reasonably well structured, so automating batch runs is possible if you are willing to do the work.
What It Is Good For
Teaching introductory physics concepts. The visual clarity of seeing forces and motions play out without graphical distractions makes it effective for classroom demos. Creating quick prototypes of mechanical interactions before building something in a heavier engine. Testing whether a basic linkage or spring arrangement behaves as expected before committing to production code. I used it extensively to validate the behavior of a series of spring-mass-damper configurations for a personal project. Setting up each variant took roughly five minutes compared to the half hour it would have taken in Unity or Unreal with their physics plugins. The tradeoff is that once the simulation proved what I needed, I had to re-implement the same behavior from scratch in the target engine using equivalent parameters. The translation is usually straightforward since the underlying math maps directly, but it is an extra step you should factor into your timeline.
Practical Tips From Using It Day to Day
Save your scenes frequently. The app does not auto-save and a crash mid-edit means losing whatever you were working on. The file format is plain JSON so corruption is rare but has happened to me twice when the app terminated during a write operation. Use the undo history liberally. It goes back about thirty steps, which covers most editing sessions but can feel short if you are iteratively refining a complex setup. I learned to commit major changes as separate scene files rather than relying on undo to bridge large revisions. If you are running on integrated graphics, lower the render resolution before adjusting the physics time step. The two are decoupled in the rendering pipeline but both compete for CPU resources on weaker hardware. I found that dropping the viewport resolution from native to half resolution kept the simulation stable at higher iteration counts without visually degrading the output enough to matter for my purposes.
The debugger overlay showing velocity vectors and contact normals is useful but toggles it on and off frequently rather than leaving it on permanently. It adds overhead and can make the viewport hard to read in dense scenes. My habit became turning it on only when I needed to verify a specific interaction, then turning it off once I confirmed the behavior was correct.
Alternatives Worth Knowing
If you need concave mesh support or continuous collision detection, PhET Interactive Simulations from the University of Colorado offers browser-based physics tools with broader shape support, though they are more constrained in terms of customization. For a full desktop experience with scripting, Algodoo remains the closest alternative and supports Lua-based automation. The tradeoff is that Algodoo has accumulated feature bloat over many years and its interface is significantly more cluttered than Minimalist Physics. For web-only projects where you do not want to distribute a desktop app, the Matter.js library gives you programmatic control over similar physics behaviors with modern browser support. It requires coding knowledge but eliminates the need for any download or installation on the user end. The choice depends on what you actually need. If you want something that opens, lets you drag objects around, and simulates cleanly without any setup overhead, this is it. If you need deeper automation or more complex geometry, you will outgrow it quickly and should look elsewhere.
