AI Tea Party Overview

A local multi-character roleplay workspace, with a verified first room and archive workflow.

What the product does

AI Tea Party is a developing web application for multi-character roleplay. A room combines characters, a user persona, world information, variables and conversation history. A Pi Agent-based backend can schedule dialogue and use tools to write messages, update the room bar, ask the user a question, or patch an existing message.

The documented public snapshot uses a TypeScript/Fastify backend, SQLite and a Next.js frontend. Older Python-backend instructions do not describe this version. The browser currently displays the name “Index Rerum.”; that is the interface in this snapshot, not a different installation.

A first result without a model key

Quick start gets you from a pinned source checkout to a running page, a room variable and a persistent archive. Templates then creates a reusable scene setup from an actual room archive. These operations were verified on 2026-09-06 with Node 22.23.2 and pnpm 10.12.1.

The local API flow created a room, character and variable, retrieved an archive and exported a template. Browser verification separately confirmed a variable and archive in the default room. Live model dialogue, OAuth login and desktop packaging were not part of that verification.

Data and deployment boundary

Room data lives in SQLite; archives are JSON files on disk. Model operations can send room context to a configured provider. Credentials may come from application settings, environment variables or the existing Pi login file. A local database does not make model processing local.

Use the guide's loopback-only development commands. This snapshot's REST routes are not a documented multi-user authentication boundary; do not treat the local development server as a ready public hosted service. The source manifest declares MIT, while this snapshot contains no standalone license file. These guides describe the public source and do not grant additional rights to its code.

Read in order

Quick start: install, open the page, create and verify an archive.
Rooms and dialogue: characters, variables, model actions and Ask.
Templates: produce a reusable configuration and understand what the exporter omits.
Troubleshooting: processes, paths, connection settings and output checks.