nanobot: an ultra-lightweight self-hosted AI agent runtime with multi-channel chat integration and persistent memory

What it solves

nanobot is a self-hosted, ultra-lightweight personal AI agent runtime. It solves the problem of needing a complex, monolithic platform to run a personal AI agent that can maintain persistent workflows, remember long-term context, and interact across multiple chat platforms while remaining fully owned and customizable by the user.

How it works

It centers around a small agent loop where messages are received from various chat channels, and an LLM decides when to trigger specific tools. Instead of a heavy orchestration layer, memory and skills are pulled in as context. The system supports model routing with fallbacks and can be deployed as a long-running gateway on a local machine or server.

Who it’s for

It is designed for users who want a private, self-hosted AI agent that can be integrated into their existing chat apps (like Telegram, Discord, or Slack) or used via a WebUI, and for developers who want a readable, extensible codebase to build personal automations.

Highlights

• Multi-channel reach: Connects to Telegram, Discord, Slack, WeChat, Email, Mattermost, and Feishu.
• Model flexibility: Supports OpenAI-compatible APIs, local LLMs (via Ollama or vLLM), and image generation.
• Tool integration: Includes built-in tools for shell access, web search, web fetch, MCP (Model Context Protocol), and cron jobs.
• ရေး-long-term memory: Uses "Dream" to maintain session history and long-term memory.
• Developer friendly: Provides a Python SDK and an OpenAI-compatible API for external integrations.