# CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

## Project Overview

DevDen is a self-hosted AI chat platform that enables organizations to provide AI-powered Q&A based on their own knowledge bases. Users interact through a clean chat interface while administrators manage knowledge bases, AI providers, and user access through a terminal-style dashboard.

**Current Status:** Early prototype stage with static frontend (HTML/CSS/JS). Backend infrastructure is not yet implemented.

## Architecture

**Planned Technology Stack:**
- **Frontend:** Vanilla JavaScript (prototype), potential migration to Svelte
- **Backend:** FastAPI (Python)
- **Database:** PostgreSQL (users, conversations, settings)
- **Cache:** Redis (sessions)
- **Vector DB:** Qdrant or ChromaDB (semantic search)
- **Auth:** Microsoft Entra ID
- **AI Providers:** Claude, OpenAI, Gemini, OpenRouter, Ollama (local)

**Two-Interface Design:**
- Clean chat interface for end users (warm, accessible design)
- Terminal-style dashboard for administrators

## Current Files

- `index.html` - Main chat interface structure
- `script.js` - Chat interaction logic (mock responses currently)
- `style.css` - UI styling with fox-themed color scheme (accent: `#d4a574`)

## Development Phases (from project.md)

1. Core Chat - Basic chat with single AI provider
2. Knowledge Base - Document indexing and semantic search
3. Authentication - Microsoft Entra ID integration
4. Admin Dashboard - Management interface
5. Multi-Provider - Multiple AI provider support
6. Git Integration - Auto-sync from repositories
7. Polish - Production readiness

## Commands (Future)

When Docker infrastructure is implemented:
```bash
docker-compose up -d
docker-compose exec devden-web python manage.py init-db
docker-compose exec devden-web python manage.py create-admin --email admin@company.com
docker-compose logs -f devden-web
```

## Design Decisions

- Knowledge bases connect via Git repositories (auto-sync) or Docker volumes (local files)
- Source transparency: answers show which documents were referenced
- User quotas configurable per-user (questions per day/month)
- API keys encrypted at rest
- All data stays on-premises (self-hosted)