microbot

Lightweight AI Assistant for Vietnamese Communities

What is microbot?

microbot is a lightweight personal AI assistant built on top of two open-source projects:

• nanobot by HKUDS — ultra-lightweight agent framework (~4,000 LOC) with multi-channel support
• kioku-agent-kit-lite by phuc-nt — tri-hybrid memory system (BM25 + vector + knowledge graph) in a single SQLite file

We take nanobot's minimal agent architecture and enhance it with kioku's persistent memory system, then add integrations focused on the Vietnamese market.

Why microbot?

	nanobot	microbot
Core architecture	~4,000 LOC	Slightly more, still minimal
Memory	Basic session history	Tri-hybrid persistent memory (kioku)
Target audience	General / Chinese communities	Vietnamese communities
Integrations	WeChat, Feishu, DingTalk...	Zalo, Vietnamese payment, local services
Philosophy	Ultra-lightweight	Lightweight + easy to maintain

Key Features

• Multi-channel: Telegram, Discord, Slack, Web UI, Zalo, and more
• Multi-provider: OpenAI, Anthropic, Google, DeepSeek, local models via LiteLLM
• Persistent memory: Long-term agent memory powered by kioku's tri-hybrid search
• MCP support: Extend capabilities with Model Context Protocol tools
• Scheduled tasks: Cron-based automated agent actions
• Vietnamese-first: Optimized prompts, integrations, and UX for Vietnamese users

Quick Start

Prerequisites

• Python >= 3.11
• Node.js >= 18 (for web UI)

Installation

git clone https://github.com/Luvu182/microbot.git
cd microbot

# Install Python dependencies
pip install -e .

Run

# Start the gateway (API + Web UI)
microbot gateway

Open http://localhost:18790 in your browser to access the Web UI.

Configuration

All configuration is done via Web UI or ~/.microbot/config.json:

# Or copy the example config to get started manually
cp config.example.json ~/.microbot/config.json

See config.example.json for all available options.

Project Structure

microbot/
├── nanobot/           # Core agent engine (from nanobot)
│   ├── agent/         # Agent logic & tool execution
│   ├── channels/      # Chat platform integrations
│   ├── providers/     # LLM provider adapters
│   ├── session/       # Conversation & memory management
│   ├── skills/        # Agent skill system
│   └── config/        # Configuration management
├── bridge/            # Node.js bridge for platform APIs
├── web/               # Web UI (Vite + React)
├── tests/             # Test suite
└── docs/              # Documentation

Supported Channels

Channel	Status
Web UI	Supported
Telegram	Supported
Discord	Supported
Slack	Supported
Feishu / Lark	Supported
DingTalk	Supported
WhatsApp	Supported
Email	Supported
QQ	Supported
Matrix	Supported
Zalo	Planned

Supported Providers

Any model supported by LiteLLM, including: OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Mistral, Qwen, local models (Ollama), and more.

Contributing

Contributions are welcome! Please open an issue or submit a pull request.

Credits & Acknowledgments

This project builds upon the work of:

• nanobot by HKUDS — The core agent architecture. nanobot proved that a fully-featured AI assistant can be built in ~4,000 lines of code.
• kioku-agent-kit-lite by phuc-nt — The memory system architecture. kioku's tri-hybrid search (BM25 + vector + knowledge graph) provides persistent, intelligent memory with zero external infrastructure.

Both projects are MIT licensed. We are grateful for their contributions to the open-source AI community.

License

MIT — Same as the upstream projects.