Build Your Own OpenClaw from Scratch with Ivan Leo (Manus)
Coding agents are everywhere now, but most of them are black boxes. In this session, you’ll build one from scratch.
Join Ivan Leo (Manus) and Hugo Bowne-Anderson (Vanishing Gradients) for a live, hands-on build. Starting from a blank Python file, we'll construct an OpenClaw-style agent that lives on your machine and responds via Telegram.
You send your agent a voice message. It doesn't have Whisper installed, so it tells you "give me a sec," downloads Whisper, and transcribes it. You ask it to turn a markdown file into a PDF. It installs Quarto and renders it. The agent figures out what it needs and gets it. You don't download extensions or skills. You ask the agent to extend itself.
What we'll cover
From Zero to Agent: Build an agentic loop from first principles: LLM calls, tool use, reasoning loops, conversation management
Sandboxed Execution: Add persistent code execution so your agent can install packages, write files, and maintain state across sessions
Telegram Integration: Connect your agent to a messaging interface you can talk to from your phone, anywhere
Self-Extending Skills: Implement a skills system so your agent can search the web, download tools, and install its own capabilities on the fly
Persistent Memory: Give your agent a to-do list, a journal, or any state it needs to remember what you've asked it to do
Along the way, we'll connect it to Gmail, Google Calendar, web search, and code review to show how the skills system works in practice.
Whether you're an ML engineer curious about agents, a developer who wants to understand the OpenClaw hype, a founder thinking about agent infrastructure, or someone who just wants a personal AI assistant that actually does things, this session will help you build.
Register to join live or get the recording afterwards.