In this episode, Toby is joined by Xavier (Zavi) for a relaxed conversation about OpenClaw, an open-source project that lets you build a personalised, memory-aware AI assistant running on your own hardware. They share hands-on experiences setting it up with Telegram, Claude and local models, and discuss what makes it feel different from a standard chat interface: persistent memory in markdown files, heartbeat schedules, proactive check-ins, and a soul file that shapes personality over time. The conversation also covers security, prompt injection risks, the skill ecosystem, local model options, and the cultural questions around long-running AI companions.

Chapters

0:00 Introduction

0:33 What is OpenClaw?

1:40 Why does it feel different from a standard AI chat?

3:51 Setting it up: first impressions

4:45 Practical use cases: standups, workshop manuals, tractor parts

7:04 How the heartbeat and memory systems work

9:15 Cron jobs, proactive tasks and the soul file

12:06 The internals: TypeScript, service daemon, CLI and web UI

14:23 Security model: token auth, Tailscale, least-privilege access

17:42 Prompt injection risks

21:30 The skill ecosystem and supply chain risks