Foxpit

Multi-agent discussion platform where multiple LLMs discuss a topic across rounds, with a user-designated Leader who summarizes each round and produces a final consensus, and a First Responder who speaks first each round. All agents are neutral — no pre-assigned positions.

Runs as both a standalone web app and a VS Code extension from a shared codebase.

How it works

1. Pick a topic and select 2+ LLM models (Claude, GPT-4o, Gemini, etc.)
2. Assign one model as Leader (summarizes rounds + writes consensus) and one as First Responder (speaks first)
3. Choose number of rounds and provider mode (OpenRouter for live calls, or Demo mode)
4. Watch the models discuss — each round ends with a leader summary, and the final round produces a consensus

All LLM calls are routed through OpenRouter — one API key gives access to models from Anthropic, OpenAI, Google, and more.

Project structure

packages/
  shared/     Core logic: discussion engine, LLM providers, types
  ui/         Platform-agnostic UI: setup screen, discussion screen, styles
  web/        Standalone web app entry point
  extension/  VS Code extension entry point

• shared — async generator discussion engine, OpenRouter provider, mock provider, type definitions
• ui — DOM rendering + adapter interface (no framework, no React)
• web — runs the engine in-browser, settings stored in localStorage
• extension — engine runs in the extension host, webview renders the UI via postMessage bridge

Setup

npm install
npm run build

Web app

npm run dev:web

Opens at http://localhost:3000.

VS Code extension

1. Open the repo in VS Code
2. Press F5 to launch the Extension Development Host
3. In the new window: Cmd+Shift+P → "Foxpit: Start Discussion"

Configuration

Web app

Select "OpenRouter (Live)" in the provider toggle and enter your API key. Settings persist in localStorage.

VS Code extension

Settings are available under foxpit.* in VS Code settings:

• foxpit.providerMode — "openrouter" or "mock" (default: "mock")
• foxpit.openrouterApiKey — your OpenRouter API key

Or use the in-panel provider toggle — settings sync to VS Code config automatically.

Available models

• Claude Sonnet 4 (anthropic/claude-sonnet-4)
• Claude Haiku 4.5 (anthropic/claude-haiku-4.5)
• GPT-4o (openai/gpt-4o)
• GPT-4o Mini (openai/gpt-4o-mini)
• Gemini 2.5 Flash (google/gemini-2.5-flash)

Architecture

The discussion engine (packages/shared/src/engine.ts) is an async generator that yields messages one at a time. The UI consumes them via an adapter interface — the web app implements it directly, the extension implements it over postMessage. The engine and UI are fully decoupled.

[Setup UI] → DiscussionConfig → [Adapter] → [Engine + Provider] → yields Message → [Adapter] → [Discussion UI]

License

MIT