Summary

This PR adds two new community plugins to PyGPT:

🎭 Plugin: Avatar Ready Player Me (avatar_rpm)

A floating 3D avatar window powered by Ready Player Me (free tier) and Three.js, rendered inside PyGPT's existing WebEngine — no additional renderer overhead.

Features:

• 🔊 Lip-sync reactive to audio output chunks (via on_audio_output_chunk hook)
• 😊 Emotion detection based on text sentiment (Spanish + English, TextBlob + keyword matching)
• 🌀 Idle animations (breathing, subtle rotation)
• 🪟 Floating window, always-on-top, fully configurable size/position
• 🎨 One-click avatar creator via Ready Player Me web
• Morph target-based expressions: happy, sad, surprised, angry, thinking, neutral

Overhead: ~15 MB RAM, <1% CPU idle, ~4% CPU when speaking.

🧠 Plugin: MCP Memory Cloud (mcp_memory)

Persistent cloud memory for PyGPT using Firebase Firestore (free Spark tier) exposed as an MCP server.

Features:

• 💾 Auto-saves conversations to Firestore in realtime
• 🔍 Semantic search with local embeddings (sentence-transformers/all-MiniLM-L6-v2) — no API key needed
• 🌐 Multi-device sync — resume context on any machine
• MCP tools: remember_fact, semantic_recall, summarize_context, sync_devices
• Supports local / openai / gemini embedding backends
• Standalone MCP server (mcp-memory-server/server.py) via stdio transport

Free tier Firebase limits (more than enough for personal use):

• 1 GB storage, 50K reads/day, 20K writes/day

Overhead: ~50 MB RAM (with embeddings loaded), <1% CPU idle.

Files Added

src/pygpt_net/plugin/avatar_rpm/
  __init__.py          # Plugin entry + hooks
  avatar_window.py     # QWidget with WebEngine + Qt WebChannel bridge
  lip_sync.py          # Audio-reactive lip-sync controller
  emotions.py          # Sentiment-based emotion detector
  assets/
    avatar_viewer.html # Three.js viewer with RPM GLB loader

src/pygpt_net/plugin/mcp_memory/
  __init__.py
  plugin.py            # Plugin entry + MCP server lifecycle

mcp-memory-server/
  server.py            # MCP server (stdio transport)
  firebase/client.py   # Firestore CRUD wrapper
  tools/remember.py    # remember_fact tool
  tools/recall.py      # semantic_recall tool (embeddings)
  tools/summarize.py   # summarize_context tool
  tools/sync.py        # sync_devices tool
  requirements.txt
  config.example.json

install.sh             # One-command install script
README.md              # Full documentation

Setup Requirements

# MCP Memory Server
pip install firebase-admin sentence-transformers numpy textblob

# Avatar RPM
pip install textblob numpy
# (WebEngine and Three.js are already included in PyGPT)

Notes

• Both plugins are opt-in and disabled by default — zero impact on existing users
• The MCP Memory server runs as a subprocess, isolated from PyGPT core
• Avatar uses PyGPT's existing QWebEngineView — no new Qt dependencies
• Firebase free tier is sufficient for personal/single-user usage
• Fully compatible with existing PyGPT plugin hooks (on_audio_output_start/chunk/end, on_ctx_begin/end)

Developed as a community enhancement. Feedback and improvements welcome!