Quick start

You will first need to register an account.

Note: For now the only way to open an account is to contact us directly for a trial.

Once you have an account you will be able to log in to Prorok Portal at portal.prorok.ai, in the supplied native Portal application or via the command-line tool.

Portal is our core visual tool enabling you to make predictions, manage and view live simulated worlds, snapshots and replays. You can also manage your account directly from Portal, including setting up compute, projects, and more.

Assembling compute

To be able to get started you will need to add at least one compute node.

Note: The option to readily use on-demand cloud compute is currently on the roadmap.

If you're running the native Portal application you will be able to make use of the local node automatically started by the Portal executable.

If you're running Portal in the browser you might need to add your first node manually, unless it was already provisioned for you with your account.

Go for examples

The fastest way to get started is to open one of the available examples from Portal's welcome screen.

Upon clicking on any of the examples, you will be taken to a new workspace. Explore the newly created world, step it forward, inspect the various entities in the world and see what they are attempting to accomplish over time.

Start prompting

You can jump straight into things by asking the prorok agent about any of the examples. Once you're in a workspace with a specific example world initialized, you can also ask the agent to tell you more about the example and even to modify the workspace for you.

Once you've familiarized yourself with the example, you can jump to examining it more deeply. Consider the boids example.

> How does predators' chances
of catching a bird change with
flock cohesiveness? 

The agent will start working on the answer. It will register a new experiment, defining the number of runs, targetted variable mutations across runs, etc. With that done it will proceed with running the experiment and giving you an answer once it's done.

You can make your query more specific as to how you're expecting the agent to proceed with their analysis.

> ... Perform a granular search over the
flock cohesiveness set of variables,
performing at least 1000 simulation runs
total.

Similarly you can instruct the agent on how it should display the results.

... Visualize with two replays best
illustrating the answer, 3d views, side
by side, each with it's own timeline
controller.

Predict

Analyzing potential causations and correlations in abstract terms is great, but what about actually predicting future states of a particular world?

Well, it's the same process, but for the latter you will need to focus on a particular starting state.

Example starting states tend to be "boring", as they are usually programmatically setting up entities in the world. You're much better off running a particular world for some time and then attempting to ask questions about the possible futures. Alternatively, look for scenarios and/or snapshots provided alongside the model.

Try initializing a world from the boids example and stepping it a few hundred ticks. Find a predator entity. Now prompt the prorok agent with the following.

> Is the predator_03 going to catch any
birds before it runs out of stamina?

You should get a simple yes-no answer based on performed simulation runs. You can always ask for more details on the actual basis for the answer, get more concrete probability values, even ask for visualization of certain outlier runs.

You can continue to prompt the agent to, for example, find possible futures where small change to the environment makes a large effect on the outcome.

Command-line

For all the power users out there, note that you can always choose to prompt the agent directly from the command line using Prorok CLI. The following command will open an interactive prompt interface in your favourite terminal interface.

prorok agent

Our CLI comes included with Portal, meaning you can ask the agent to visualize anything you're working on.

Wider AI integration

A powerful feature of Prorok AI is the fact that it can be used with any commercial LLM agent.

Note: We currently do not provide an MCP interface, so for now you are limited to AI agents capable of executing CLI applications.

Start working on your own model

You've familiarized yourself with Portal. Likely you're looking to gain insights into some particular system. You will first need to create a model of that system.

See the relevant how-to page for a complete guide to creating Prorok models from scratch.