Vensim Help

AI Assistant

The AI Assistant is a conversational helper that opens in a pane beside your model. It can explain how a model works, help track down an error, suggest new structure, teach system dynamics from scratch, and write up a conversation as a shareable document.

The Assistant reads the model you have open and stays in step with it. When you run a simulation, check units, check the model, or add, rename or delete a variable, Vensim passes the result across automatically — there is nothing to copy and paste.

You supply your own AI provider account, so what you spend is billed directly by that provider. See Connecting an AI provider below.

Opening the Assistant

Click the AI Assistant button on the Vensim toolbar. The Assistant opens in a pane beside your model, and the model you are working on loads into it automatically.

The first time you open it you will be asked for an API key before it can answer anything. Once entered, the key is remembered and you will not be asked again.

Closing the pane does not lose your conversation — reopening it picks up where you left off.

Connecting an AI provider

The Assistant needs an API key before it can answer. Open the settings panel on the right of the pane, choose your provider, and paste the key into the API key field. The question-mark button beside the field gives step-by-step instructions for the provider you have selected, including where to create a key.

Ollama and LM Studio run a model locally on your own computer. Nothing is sent over the internet, and no key or account is needed, but you do need the local application installed and running first.

Your key is held in your browser's local storage on your own machine. It is sent only to the provider you chose, and to no one else. Remove key deletes it again.

The three modes

Buttons at the top of the pane switch between three modes. You can move between them at any time; the conversation carries over.

Learn

A guided walkthrough that builds a working model with you from nothing, assuming no prior experience. It runs in three stages, shown in a tracker at the top of the pane:

•Design — on paper, no software. You settle what behaviour you are trying to explain, sketch how the graph should look over time, and agree what accumulates and what makes it rise and fall.

•Build — you build the model in Vensim. The Assistant tells you exactly which tool to pick and what to type, one step at a time.

•Test — you run the model and compare the result against the behaviour you predicted in Design.

You do the building yourself throughout; the Assistant guides rather than takes over. It will not move you on to the next stage until the current one is genuinely finished.

Chat

Ask anything about the model in plain language. There is a single conversation for every kind of question — explaining how a structure produces the behaviour you are seeing, working out why an error appears, or brainstorming what to add next. The Assistant works out what you are asking for and adapts; a small flag beside the mode buttons shows what it is currently doing.

Tools appear in the conversation when they become relevant, such as sketching the behaviour you expect, or summarising the discussion as a document.

Doc

Turns the conversation into a short, shareable write-up. The Assistant asks a few questions — who the document is for, which variables matter most — and assembles a summary section by section, including a causal loop diagram built up as it goes. The result can be exported and circulated.

What the Assistant knows about your model

The open model loads into the pane by itself. A summary appears at the top of the conversation showing:

•how many stocks, flows, auxiliaries and parameters the model contains;

•the simulation start and end times, the time step and the time units;

•how many feedback loops were found;

•any unit problems;

•model-quality advisories — for example a number written directly into an equation where a named constant would be clearer, or a stock starting at a value other than zero without a constant behind it.

These advisories are suggestions about model style, not errors. The model still runs.

Use Reload to pull the current state of the model across again after editing it in Vensim.

What Vensim sends across automatically

While the Assistant is open, Vensim forwards these events as they happen, so you can simply ask about what just occurred:

•the results of a simulation run, including any errors or warnings;

•the output of a units check;

•the output of a model check;

•a variable being added, renamed or deleted;

•floating-point errors reported during SyntheSim.

The Assistant can also ask Vensim for things it needs mid-answer, such as the time series behind a variable or its value at a particular time, so it can discuss what the model actually produced rather than what it ought to produce.

The structure and loops panel

A panel on the side of the pane lists the feedback loops found in the model, each marked Reinforcing or Balancing, with the chain of variables that forms it. Where a loop's polarity cannot be settled from the structure alone it is shown as ambiguous.

Molecules of Structure

A built-in reference library of 68 canonical system dynamics patterns from Jim Hines' Molecules of Structure. Each entry shows the diagram, the equations and a description of when the pattern applies. Open it from Tools > Molecules.

The Assistant also points to a molecule by name during a conversation when the structure under discussion matches one.

Sessions and history

Every message is saved to your browser as you go. You can close the pane or Vensim and pick the conversation up where you left it.

History lists your earlier sessions so you can reopen one. From there you can also:

•Download backup — save a session as a file;

•Restore from file — reopen a saved session, including on a different computer;

•Delete — remove a session.

Exporting a conversation

Export saves the conversation in either of two formats:

•Markdown — plain text, suited to documentation tools and version control;

•HTML — opens in any browser, and can be printed to PDF.

Appearance

A− and A+ scale the whole pane between 75% and 150%, which helps when the pane is docked narrow. The sun and moon button switches between light and dark; this follows the Vensim theme by default.

When the pane is narrow, the text size controls, the Molecules library, History and Export move into the ⋮ menu in the top right.

Reporting a bad answer

Each reply carries a small flag button. Clicking it sends that exchange to the Vensim team so the Assistant can be improved. The dialog shows exactly what will be included before anything is sent, and you can choose whether to attach the full conversation. Nothing is sent unless you click the flag.

Taking the tour

Tour runs a short walkthrough of the pane, under a minute. It can be restarted at any time from the same button, or from the ⋮ menu when the pane is narrow.

What leaves your computer

Your messages and the parts of the model needed to answer them are sent to the AI provider you selected, in order to produce a reply. Your API key is sent only to that provider. If you use Ollama or LM Studio, nothing leaves your machine at all.