About the System One Playground

The System One Playground is a browser-based workbench for testing decision models. Describe a state, then ask a choice, yes/no, or scored question. Each run returns the model’s answer, the probability distribution behind it, latency, token counts, and the exact JSON response, so you can inspect how a prompt behaved instead of judging it by a single label.

It connects directly to a local Ollama instance running Nimble or Clef Flash, or to TypeSafe’s hosted Jev API. Local models are called from your browser; hosted models use a TypeSafe API key. Connection settings and sessions are stored in this browser’s local storage, and images attached to a request stay in the tab.

How to use it

1. Open Connection settings and choose a local Ollama model or TypeSafe Jev. Test the connection before starting a session.
2. Describe the state the model should reason about, then add one or more questions. Choice questions take a list of options; scoring questions take ordered levels.
3. Run the session and compare the verdict, confidence, latency, and token counts. Past runs stay in the sidebar, and the Response and Request tabs show the exact JSON that crossed the wire.

The playground is a diagnostic tool, not a production API. Use it to compare models on the same task, test prompt wording, or inspect a response before integrating a model into an application.