Ronginus - AI Debate Plugin for GemiHub

Ronginus is a shared GemiHub and GemiHub Desktop plugin that enables structured debates between multiple AI participants with different roles and models. Participants discuss a given theme across turns, draw conclusions, and vote for the best one.

Features

• Multi-model Debates: On GemiHub Desktop, choose a configured OpenAI-compatible, Gemini, Vertex AI, Anthropic, or local-agent model for each participant.
• Role-based Discussion: Assign roles (e.g. "Affirmative", "Critical") to each participant. The same model can participate multiple times with different perspectives.
• User Participation: Join the debate yourself alongside AI participants.
• Turn-based Discussion: Configurable number of turns (1-10). Each participant sees all previous responses before responding.
• Conclusion & Voting: After discussion turns, each participant provides a final conclusion, then all participants vote for the best one.
• Transcript Export: Save the complete debate transcript to Google Drive on Web or the active Workspace on Desktop.
• Persistent Settings: System/conclusion/vote prompts are saved per-plugin via the storage API.
• i18n: English and Japanese (auto-detected from browser locale).

Installation

Via GemiHub Settings or GemiHub Desktop 0.8.1+

1. Open GemiHub Settings > Plugins tab
2. Enter takeshy/hub-ronginus and click Install
3. Enable the plugin

Both hosts use the same GitHub Release. GemiHub loads main.js; GemiHub Desktop applies the repository-owned patches/gemihub-desktop.patch and saves transcripts in the active Workspace.

Building from Source

git clone https://github.com/takeshy/hub-ronginus
cd hub-ronginus
npm install
npm run build

This produces main.js, styles.css, manifest.json, and patches/gemihub-desktop.patch for a GitHub Release.

Screenshots

Setup	Discussion	Voting	Result

Usage

1. After installation, the AI Debate panel appears in the right sidebar.
2. Enter a debate theme/topic.
3. Set the number of turns.
4. Add participants — choose AI or User and optionally set an AI model and a role. Desktop lists configured models; leaving the model blank uses the host's current model.
5. Click Start Debate.
6. AI participants generate responses sequentially. If you added yourself as User, you'll be prompted for input.
7. On the final turn, each participant provides a conclusion.
8. All participants vote for the best conclusion.
9. Winner (or draw) is announced.
10. Click Save to Drive on Web or Save to Workspace on Desktop to export the transcript.

Settings

Expand the Settings section on the debate panel to customize:

Setting	Description
System Prompt	Base instructions given to all AI participants
Conclusion Prompt	Prompt appended on the final turn for conclusions
Vote Prompt	Prompt for the voting phase (vote format instruction is auto-appended)

Plugin API Usage

This plugin uses the following GemiHub Plugin APIs:

• api.registerView() — registers the debate panel as a sidebar view
• api.llm.listModels() (Desktop) — lists providers and models configured in the host
• api.llm.chat() (Desktop) / api.gemini.chat() (Web) — sends messages to the participant's selected model with role-specific system prompts
• api.drive.createFile() — saves debate transcripts to Google Drive
• api.storage.get/set() — persists plugin settings

How It Works

Theme Input + Participant Selection
    |
    v
+------------------------------------+
|  Turn 1                            |
|  AI(Model A, Role A) -> AI(Model B, Role B) |
|  -> User(Role C) -> ...            |
+------------------------------------+
    | (each sees previous responses)
    v
+------------------------------------+
|  Turn 2 ~ N-1                      |
|  Refined thoughts based on         |
|  previous discussion               |
+------------------------------------+
    | (final turn = conclusion)
    v
+------------------------------------+
|  Final Turn                        |
|  Each participant provides a       |
|  final conclusion                  |
+------------------------------------+
    |
    v
+------------------------------------+
|  Voting Phase                      |
|  All participants vote for the     |
|  best conclusion                   |
+------------------------------------+
    |
    v
Winner Announced (or Draw if tied)

License

MIT License