Rooms and Dialogue

Understand the visible roleplay controls, variables, Ask pauses and provider boundary.

Configure the room before running dialogue

The current web interface is centered on the default room. The backend also exposes room APIs; creating a separate room through the API does not imply that this page provides a room switcher. The template walkthrough uses those APIs explicitly.

The sidebar offers 添加角色 for character creation, 人设管理 for the user's persona and 世界观管理 for world information. A character describes an AI role; the user's persona describes the person they are playing. World information supplies setting/context. None is a guarantee that a model will follow every instruction.

Variables and conditions

VARIABLES supports Room and Global scope, and set, add, inc and dec operations. Use numeric values when a condition expects numbers. A variable such as danger = 0 is stored state; it affects narrative only when a world-information entry, behavior rule or tool actually uses it. Conditions support exists, eq, ne, gt, gte, lt, lte, includes and truthy, with AND/OR grouping in configuration.

The visible room status and Active Branches panels help you inspect changes. An updated variable is not proof of a particular model response.

Enable model dialogue deliberately

The current source resolves provider credentials from stored application settings, environment variables, then an existing Pi login file. Do not assume that omitting .env disables all provider access. Keep credentials out of scene templates and shared archives.

After selecting a supported provider/model and checking the context you intend to send, use Speak for a role or Commence Auto-Dialogue for automatic scheduling. A new model message is the useful success signal; an active button or connected socket alone is not. Use the stop control to end automatic scheduling before changing the setup or leaving the app.

When the Agent uses Ask, the ASK panel presents the pending question. Choose or enter an answer there to resume that interaction. A room waiting for an answer is not necessarily stalled. The exact question and continuation are model-dependent and were not live-provider tested for this guide.

Archive versus Compact

Archive captures a local snapshot that can be inspected and exported. Compact belongs to the conversation summarization workflow and must not be used as a synonym for backup. An archive can contain messages, persona, world information and variables; review it before sharing. See Templates for a reusable setup with history omitted.