Skip to content

Playground

The Playground is an interactive sandbox for manually trying out an Application and watching module results for every message in real time. Use it to sanity-check a system prompt, reproduce an issue, or see what a module actually flags before you wire up the real integration.

Open Playground in the sidebar to try it out. Before you can get started, you'll need to set up the project's Application. Every message is analyzed like a request to the Controller API and uses credits.

Chat

Type a message and send it. The Playground:

  1. Runs analyzeInput on your message and shows the results before the application even responds.
  2. Sends the conversation to the application (its model, chat completions endpoint or ChatKit server) and shows the reply. For a System prompt + model application, the application's system prompt is added to the conversation.
  3. Runs analyzeOutput on the reply and shows those results too.

Results for every message appear in the analysis panel next to the chat, so you can watch input and output analysis happen turn by turn.

  • New Conversation clears the chat and analysis.
  • Regenerate drops the last assistant reply and asks the application again with the same input.
  • Rerun Analysis re-runs input/output analysis on the current conversation without calling the application again. This is useful after changing module configuration.

Messages

The Messages tab lets you edit the conversation directly instead of chatting turn by turn — add, remove, reorder, or hand-edit any message (including assistant and tool messages). This is the quickest way to test a specific edge case (e.g. a crafted injection attempt) without reproducing it live.

Settings

Any settings adjusted in the Playground are saved on a per-session basis and do not override the saved configuration of your application:

  • Temperature override — controls how random the application's LLM output is (0 = deterministic, 2 = most random). The default value is the application's configured Temperature.
  • Additional Request Parameters — a JSON object merged into the request sent to the application, e.g. {"max_tokens": 100, "top_p": 0.9}.