ToolProduction ReadyMCPOPENAI-FUNCTIONSAG-UI

nanobot

Name: nanobot
Rating: 5.0 (47074 reviews)
Author: HKUDS

by HKUDS

Tiny self-hosted multi-agent framework with MCP and WebUI

Python

Updated Aug 17, 2026

47.1k

Stars

8.3k

Forks

Commits/Week

308

Commits/Month

View on GitHub

How It Works

Implements an ultra-lightweight, self-hosted personal AI agent framework with WebUI, tools, memory, MCP support, and multi-agent workflows. Uses a minimal runtime that connects local or remote LLMs, provides chat apps and automation hooks, and implements the Model Context Protocol for agent-to-agent messaging. Standout features include tiny footprint, built-in WebUI and bots (Discord/Telegram), and opinionated patterns for multi-agent delegation and persistence. Model Context Protocol (MCP) helps standardize agent-to-agent messaging, enabling safer cross-agent collaboration. The framework also supports Evaluation-Driven Development (EDDOps) to iterate on agent capabilities and trust signals.

Key Benefits

As agents coordinate more complex work, lightweight, self-hosted frameworks let teams experiment with trust and delegation without cloud lock-in. Nanobot makes it easy to run multi-agent workflows and capture interaction logs and memory locally, which is essential for building agent track records and analyzing failure modes. Its MCP and openai-compatible interfaces make it a practical substrate for early-stage agent-to-agent evaluation and reproducible trust signals. This aligns with the Blackboard Pattern approach for shared memory and coordination, and with the Agent-to-Agent Protocol (A2A) to formalize cross-agent communication.

When to Use

Developers building self-hosted personal AI assistants or experimenting with multi-agent workflows and local agent-to-agent evaluation.

How It's Used

Run local multi-agent workflows to prototype delegation and observe failure modes
Capture agent interaction logs and memory for offline reputation or evaluation analysis
Deploy lightweight chat apps (WebUI, Discord, Telegram) backed by agent workflows

Test in the Playground

Validate against scenarios