GitHub - daegwang/atombot: A tiny but powerful personal AI assistant 🐙

2 min read Original article ↗

atombot

Atombot: Atomic-lightweight personal AI assistant

🐙 A tiny but powerful personal AI assistant inspired by OpenClaw and nanobot.

⚛️ Core functionality in just ~500 lines of code — ~90% smaller than nanobot (~4k lines) and ~99.9% smaller than OpenClaw (400k+ lines).

✨ Features

  • 🧠 Multiple LLM provider support: Supports OpenAI-compatible endpoints and Codex(CLI mode).
  • 💬 Gateway: Chat with the same agent via Telegram with allowlist-based access control.
  • 🧾 Persistent memory: Long-term memory with searchable daily history logs.
  • ⏰ Scheduled reminders: Supports both one-time and recurring reminders.
  • 🧩 Skills system: Compatible with OpenClaw SKILL.md format with metadata support.
  • 🚀 Fast onboarding: Provider-first setup that auto-detects Codex, LM Studio, and Ollama models, then bootstraps config and workspace.

📦 Installation

From source (recommended for development)

git clone https://github.com/daegwang/atombot.git
cd atombot
pip install -e .

From PyPI

🚀 Quick Start

1. Initialize workspace

The onboarding process:

  • Selects provider → model
  • Detects available providers and models (Codex / LM Studio / Ollama)
  • Bootstraps config, workspace folders, prompts, and skills
  • Optionally configures the Telegram gateway

2. Start Gateway (Telegram)

This allows you to chat with your agent via Telegram.

3. Chat with Atombot

You can interact with Atombot via Telegram or the CLI.

Telegram

  • Send a message to your configured bot.

CLI

Example:

Atombot ready. Type 'exit' to quit.
> hello
Hello! 👋 I'm Atombot, your lightweight personal AI assistant. How can I help you today?
>

The CLI provides an interactive chat interface for communicating with your agent directly from the terminal.

📁 Project Structure

atombot/
├── agent/        # agent runtime (loop, context, memory/skills/tool wiring)
│   ├── core.py   # main agent logic
│   ├── memory.py # memory read/write + recall
│   ├── skills.py # skill discovery + injection
│   └── tools.py  # local tools exposed to the model
├── prompts/      # base prompt files (AGENTS.md, MEMORY.md)
├── channels/     # chat channels (Telegram gateway)
├── provider/     # provider adapter (OpenAI-compatible + Codex CLI)
├── scheduler/    # cron storage and reminder trigger logic
└── skills/       # built-in skills (OpenClaw-compatible)