AI SvetLio Personal

A lightweight, self-hosted AI personal assistant powered by Google Gemini (or any OpenAI-compatible LLM). Runs locally with SQLite — no cloud database, no subscriptions, no telemetry.

SvetLio (from Bulgarian "Светльо", a friendly name derived from "light") is your private AI assistant that manages tasks, notes, and projects through natural conversation — via a web interface or Telegram.

Features

• Natural task detection — say "I need to finish the report by Friday" and SvetLio asks "Should I create a task with a Friday deadline?"
• Task management — create, list, update, and complete tasks with priorities and due dates
• Notes — save quick notes with tags, searchable later
• Project scaffolding — create new projects with a full memory structure (.memory/ system for AI context persistence)
• Deadline tracking — check upcoming and overdue tasks
• Web fetching — retrieve and summarize content from URLs
• Database queries — ask questions about your data in natural language (read-only SQL)
• Multi-channel — WebUI (localhost) + Telegram bot (optional)
• Multi-LLM — Google Gemini (free tier), OpenAI, Ollama, or any OpenAI-compatible provider
• Smart rate limiting — adaptive throttling based on your LLM tier (no wasted API calls)
• 66 unit tests — solid foundation with node:test

Quick Start

# 1. Clone
git clone https://github.com/SPartenev/ai-svetlio-personal.git
cd ai-svetlio-personal

# 2. Install
npm install

# 3. Configure
cp .env.example .env
# Edit .env — add your GOOGLE_API_KEY (free from https://aistudio.google.com/apikey)

# 4. Initialize database
npm run init-db

# 5. Start
npm start
# Open http://localhost:3010

Configuration

Copy .env.example to .env and configure your preferred LLM provider:

| Provider | Config | Cost | RPM | |----------|--------|------|-----| | Google Gemini API | GOOGLE_API_KEY=... | Free | 15 | | OpenAI | OPENAI_API_KEY=... | Paid | Varies | | Ollama | Just run ollama serve | Free (local) | Unlimited | | Gemini OAuth | Run gemini CLI first | Free | 1 |

Telegram (optional)

1. Create a bot via @BotFather
2. Add to .env:

   TELEGRAM_BOT_TOKEN=your-bot-token
   TELEGRAM_OWNER_ID=your-telegram-user-id

3. Run: npm run telegram

Architecture

User (WebUI :3010)  -->  server.js
User (Telegram)     -->  bot.js
         |
    session-store  -->  chat_sessions (SQLite)
         |
    skill-engine   -->  dynamic system prompt
         |
    tool-executor  -->  LLM  -->  tool-runner
         |                           |
    rate-limit queue        sqlite / web_fetch / ...
         |                           |
    LLM (next iteration)  <--  tool result
         |
    Final response  -->  User

Tools

| Tool | What it does | |------|-------------| | task_manage | CRUD for tasks — create, list, update, complete | | notes_manage | CRUD for notes — save, search, tag | | project_create | Scaffold a new project with .memory/ structure | | deadline_check | Show upcoming and overdue tasks | | db_query | Read-only SQL queries on the database | | web_fetch | Fetch and process web pages |

Database

SQLite with WAL mode. 9 tables auto-created on first run:

chat_sessions, agent_definitions, agent_tools, agent_tool_access,
agent_flows, agent_flow_steps, tasks, notes, projects

Scripts

npm start          # Start WebUI on :3010
npm run dev        # Start with --watch (auto-reload)
npm run init-db    # Initialize/reset database
npm run telegram   # Start Telegram bot
npm test           # Run all 66 tests

Tech Stack

• Runtime: Node.js 18+
• Database: SQLite via better-sqlite3 (WAL mode)
• Web server: Express 4.x
• Telegram: grammY 1.x
• LLM SDKs: @google/genai, openai
• Frontend: Vanilla JS + marked.js (CDN)
• Tests: node:test (built-in, zero dependencies)

Project Structure

ai-svetlio-personal/
├── core/
│   ├── db-adapter.js          # Database adapter factory
│   ├── db-adapter-sqlite.js   # SQLite implementation
│   ├── session-store.js       # Chat session CRUD
│   ├── skill-engine.js        # Dynamic system prompt builder
│   ├── tool-runner.js         # Tool executors (sqlite, web_fetch, etc.)
│   ├── tool-executor.js       # Agentic loop with rate limiting
│   ├── llm-router.js          # Multi-provider LLM routing
│   ├── init-llm.js            # LLM initialization from config
│   └── *.test.js              # Unit tests
├── setup/
│   └── init-db.js             # Schema + seed data
├── config/
│   ├── agents.json            # Agent configuration
│   └── prompts/
│       └── personal.md        # System prompt (reference)
├── channels/
│   ├── webui/                 # Web interface (Express + vanilla JS)
│   └── telegram/              # Telegram bot (grammY)
└── data/
    └── personal.db            # SQLite database (auto-created)

Related Projects

AI SvetLio is a family of AI assistant tools:

• ai-svetlio-personal (this repo) — Lightweight personal assistant
• ai-svetlio-office — Multi-agent office assistant (PostgreSQL, multi-user)
• ai-svetlio-brand — Brand guidelines and landing page

License

MIT