featured · github

Nanobot: A Self-Hosted Agent Framework That Doesn't Make You Build from Scratch

Open-source Python framework bundles memory, tools, and multi-agent workflows into one deployable package. Cuts weeks of scaffolding for builders shipping AI products.

HKUDS/nanobot ↗

Nanobot is a ready-to-run Python agent framework you host yourself. Think of it as the skeleton for an AI assistant that remembers context, can use external tools, and talks to other agents. Instead of wiring together five libraries and debugging state management, you get a working WebUI, persistent memory, and tool bindings out of the box. MCP support means it plays well with Claude, Ollama, and other model ecosystems. For solopreneurs, it's a launchpad for a product. For builders, it's a scaffold you extend without reinventing agent plumbing. The catch: it's battle-tested for mid-scale use, not enterprise deployment.

Share kit

Email subject

Nanobot: Agent framework minus 6 weeks of glue code

Email blurb

Self-hosted Python framework with memory, multi-agent workflows, MCP support, and a WebUI. No scaffolding tax. Open-source, extensible, ready to ship or fork. See it on GitHub.

Nanobot is what you reach for when you want to ship an AI agent without writing the agent infrastructure yourself. Self-hosted Python, MCP-native, memory and tools baked in, WebUI included. Weeks of setup time converted to hours. GitHub link in replies.

For product teams and solo builders: Nanobot removes the tedious part of building agents. Python framework, self-hosted, memory and tool management included. MCP support, multi-agent workflows, WebUI ready to deploy. The kind of scaffold that saves your first month and scales with you. Open source. Worth a look if you're shipping agent products.

Just found a self-hosted agent framework that doesn't make you build the entire stack from scratch. HKUDS/nanobot is Python, lightweight, ships with a WebUI out of the box, and handles tools, memory, MCP, and multi-agent workflows in one place. Why this matters: most agent projects are either toy examples or massive enterprise platforms. This one lets you spin up a real, extensible personal AI agent locally in minutes. Tools work. Memory persists. Multi-agent chains compose. You own the whole thing. The actual move: clone it, pip install, run the WebUI, wire up your first tool. No cloud lock-in. No vendor API overhead. Just Python and a working agent. If you've been waiting for the middle ground between "roll your own" and "pay SaaS rent," this is worth 30 minutes of your time.

nanobot is a self-hosted agent framework that actually ships complete. Python, WebUI, tools, memory, MCP, multi-agent workflows. No bloat. Runs local. You own it. If you've been DIY-ing agents and hitting the "now I need to build memory, tools, and MCP myself" wall, this saves weeks of glue code. https://github.com/HKUDS/nanobot