AI Assistant Platform — AI skill for Claude Code

AI Assistant Platform

A full-stack, multi-model AI assistant platform - chat across OpenAI, Anthropic Claude, and Google Gemini; retrieval-augmented generation over your own documents; long/short-term memory; tool & function calling; a real Model Context Protocol (MCP) server and client; multi-step agents; and integrations with Google Calendar, Gmail, Telegram, voice, and vision.

Built with **FastAPI**, **SQLite**, **Pydantic v2**, and vanilla **HTML/CSS/JS** - no frontend build step required.

---

1. Project overview

This project demonstrates an end-to-end AI engineering stack in a single, readable repository:

Layer	What it shows
Multi-model orchestration	llm_providers.py - one interface over 3 vendor SDKs
RAG	rag_service.py - chunk -> embed -> cosine search -> cited context
Memory	memory_service.py - windowed short-term + persisted long-term facts
Tool/function calling	tools.py + provider-specific tool-calling loops
Agents	agents.py - plan -> act -> observe loop with 3 example personas
MCP	mcp_server.py / mcp_client.py - real stdio MCP server + client
Integrations	Gmail, Calendar (Google OAuth2), Telegram bot, voice (Whisper/TTS), vision (GPT-4o/Claude vision)
Automation	Generic outbound webhook dispatcher for n8n / Zapier / Make

2. Architecture

Clean, mostly-flat architecture - see [`ARCHITECTURE.md`](ARCHITECTURE.md) for the full breakdown. In short:

Browser (Jinja2 templates + vanilla JS)
        │  fetch() with JWT bearer token
        ▼
FastAPI routers (routes_*.py)  ──►  Service layer (chat_service, rag_service, ...)
        │                                   │
        ▼                                   ▼
   SQLAlchemy async ORM (models.py)   llm_providers.py (OpenAI / Anthropic / Gemini)
        │
        ▼
     SQLite (data/app.db)

3. Folder structure

ai-assistant-platform/
├── main.py # FastAPI app entrypoint
├── config.py # Settings (env vars)