Loop is a browser-based system dynamics tool. Model the things that build up and drain away, the rates between them and the feedback that connects them. Run it, find the loop that is driving the result, and share what you found as a simulator anyone can use.
No install · Simulations run in your browser · loop.simmodlr.app
Customers, cash, staff, stock, infections: each one builds up over time and changes the rate at which it changes. A row of formulas can describe one step. It can't easily show you what the loop does over years.
A stock carries its history with it. Today's backlog is every order that came in minus every order that went out, so the past keeps shaping the present.
Reinforcing loops drive growth and collapse. Balancing loops hold things steady. Which one is in charge can change part way through, and that is usually where the surprise is.
When the effect of a decision arrives months later, people over-correct. Delays are built into Loop, so you can see the overshoot before it happens.
And if your question doesn't really involve feedback, Loop will tell you. Its Loop or Excel? check reads the model's equations and says whether a spreadsheet would do the job, and why.
Loop can draft a model with you, explain it back to you, and check it. Prefer your own AI assistant? Connect it.
Describe the system in plain language and review the draft before anything changes, or let a guided interview build it one question at a time.
Ask how a value is calculated or what feeds it. Answers come from the model's own structure, and asking never changes the model.
Connect an assistant such as Claude through Loop's MCP server to build, run, test and document models. It runs on your own computer over plain model files.
One model, from the first sketch to the simulator you hand to a client.
Every editor works on one shared model, so you can switch between them at any time.
Move the sliders, save scenarios and compare them side by side.
Loop finds the feedback loops, labels them reinforcing or balancing, and shows which one drove the run, and when.
Four separate verdicts instead of one score: is it sound, robust, substantiated and readable?
The classic SIR model is one of Loop's built-in templates. Three stocks of people (susceptible, infected and recovered) and two flows between them.
Run it and the feedback shows itself: infection grows on itself until the susceptible pool runs low, then the balancing loop takes over and the outbreak burns out.
Stocks are boxes, flows are pipes with valves, and clouds mark where material enters or leaves the system. Switch the view to add the auxiliaries, constants and causal links behind each flow.
The infected curve peaks as the susceptible pool runs down. Change the transmission rate on a slider and the peak moves straight away; save each setting as a scenario to compare later.
Group a model into sub-models and declare what each one takes in and feeds out. Loop describes how each part works, warns when something crosses the boundary without a port, and can test a sub-model on its own.
In a large model, pick a sub-model and see everything it reads and everything that reads it, arranged by where it lives. Download it as an image or print it.
You don't need a background in system dynamics. If the thing you care about builds up, drains away and feeds back on itself, Loop is built for it.
Build the model, run the cases, and publish them as a simulator with documentation behind it. The client moves the sliders themselves and sees the evidence for each assumption.
Recurring problems usually have a loop behind them. Map what affects what, find the loop that keeps restoring the problem, and test where a change would actually stick.
Start from a classic template, see the stocks and flows, and watch loop dominance shift during a run. Loop explains each variable from the model itself and doesn't hide how it works.
No. Describe what you're modelling in plain language and Loop drafts a first version for you to review, or start from one of the templates. The user guide explains stocks, flows and loops as you meet them.
Yes. Loop imports XMILE files, including arrays of any number of dimensions, and Vensim .mdl files without subscripts. Before the model is created it shows a report of what it changed, what it left out and what needs your review. Models also export as XMILE and JSON.
No. You can build everything by hand on the canvas or in the forms, and the model checks work with no AI configured. The AI features are there to get you started faster and to explain results.
Yes. Loop's MCP server lets an assistant such as Claude build, validate, run, test and document Loop models. It runs locally over plain JSON model files, with no account needed and no data leaving your machine. There is also a downloadable prompt pack for chat tools.
Models are saved to your Loop account and stay private unless you share them. You can invite named viewers or editors, make a model public, or publish it as a simulator. Every run keeps its own snapshot of the model, so history survives later edits.
Loop doesn't yet read data from Excel, databases or live sources (data comes in as CSV), there is no live co-editing, and integration is Euler or RK4 only. We'd rather say so than have you find out later.
Start from a template or a plain-language description.