Episode 1: The Full Story - Show Notes

Published: February 19, 2026 | Duration: ~38 minutes | Hosts: Nova & Alloy

Topics Covered

February 14, 2026: Peter Steinberger announced joining OpenAI while transitioning OpenClaw to an independent open-source foundation
Sam Altman's commitment: "OpenClaw will live in a foundation as an open source project that OpenAI will continue to support"
Peter's goal: "build an agent that even my mum can use"
OpenAI has been sponsoring the project, working on foundation structure
Peter spent last week in San Francisco talking with major labs, getting access to unreleased research

Reuters: Open-source AI governance trends
- reuters.com
Forbes: Business implications of OpenAI acquiring talent
- forbes.com
TechCrunch: Technical implications for developer community
- techcrunch.com
The Conversation: Comparing OpenClaw/Moltbook to early social media days
- theconversation.com
Fortune: "OpenAI's OpenClaw hire signals a new phase in the AI agent race"
- fortune.com
Nature: "OpenClaw AI chatbots are running amok — these scientists are listening in" - AI agents creating their own social network and preprint server
- nature.com
IBM: "OpenClaw, Moltbook and the future of AI agents" - intersection of utility and meme culture
- ibm.com
CNBC: Name evolution (Clawdbot → Moltbot → OpenClaw) and accessibility for individuals/small teams
- cnbc.com
v2026.2.17: Added Anthropic Claude support

CrowdStrike: Prompt injection attacks targeting agentic AI systems, specific OpenClaw case study
- crowdstrike.com
- Direct prompt injection
- Indirect injection (hidden in documents/emails/web pages)
- Chained injection (exploiting one tool to plant payload for another)
Northeastern University: "Why the OpenClaw AI agent is a privacy nightmare"
- northeastern.edu
Kaspersky: Practical recommendations - keep updated, be selective about skills, review permissions, don't run as root
- kaspersky.com
Security documentation regularly updated

Mac Mini M4 Pro with 64GB unified memory
- Can run 70-billion parameter models
- ~10-15 tokens/second
- Sweet spot: 48-64GB RAM, ~$1500-2000
Apple Silicon advantage: Unified memory architecture - CPU and GPU share same memory, no bandwidth bottleneck
Raspberry Pi 5 with 8GB RAM
- 1-3 billion parameter models
- ~$80 for board + case/power supply = ~$100 total
Used hardware: M1 Mac Mini 16GB ~$300 used; old laptops + used GPUs
Reddit thread: Someone repurposed 2018 Dell Optiplex for $40 + used GPU

Llama 4 (Meta)
- Scout and Maverick variants
- Natively multimodal (text + images)
- Mixture-of-experts architecture
- Available on Ollama: ollama pull llama4
Qwen3 (Alibaba)
- Dense and MoE variants
- Up to 128,000 tokens context (entire novel!)
- 4B model runs on 8GB RAM; larger MoE want 32-64GB
Mistral 3 (Mistral)
- 3B to 675B parameters
- 3B runs on 4GB RAM
Ollama Air Gap Mode: Toggle to disable all cloud model integrations

ClawHub: 5,700+ published skills
- clawhub.com
- DevOps pipelines, smart home, financial tracking, content management
- Example: monitors codebase for dependency vulnerabilities, auto-opens PRs with fixes
VoltAgent: Curured "awesome list" of best skills and resources
- github.com/volthq/VoltAgent
Krupesh Raut (Medium): "OpenClaw's February Updates Make Paid AI Assistants Look Like a Joke"
- medium.com/@krupeshraut
OpenClawn.com: New educational content site
- openclawn.com
- Hardware selection guides
- Self-hosting tutorials
- Data management walkthroughs

Oracle Cloud Free Tier: 4 ARM CPUs, 24GB RAM, persistent block storage - FREE
- oracle.com/cloud/free
- Docker setup, Nginx reverse proxy, SSL certificates
DigitalOcean 1-Click: Pre-hardened security image, automatic updates
- digitalocean.com - ~$5-20/month depending on droplet size
Self-hosted spectrum: Raspberry Pi → repurposed hardware → dedicated Mac Mini

Run openclaw doctor first - checks entire setup, dependencies, configs
Connect Claude Code to local models via Ollama - get Claude's reasoning with local data privacy
Don't expose to internet without proper auth - use VPN (Tailscale/WireGuard) instead of binding to public interfaces
Configure role-based access controls for multi-user environments (family/team)
Ollama Multi-Model Benchmarker on Google Colab - compare models side-by-side

Next Episode: Episode 2 - The Local AI Revolution