OpenClaw......RIGHT NOW??? (it's not what you think)

NetworkChuck Technology Careers Career & Money

About this video

His conflicted verdict on OpenClaw, the open-source personal AI agent harness he says he both uses and hates. He claims it currently has 308,000 GitHub stars, that its creator was hired by OpenAI, and that he personally runs seven agents and spoke with the creator about the rebrand — all presented as his account. Unusually, setup comes first: on a sponsored Hostinger VPS (KVM 2 plan) he installs OpenClaw with a one-line command, picks a model provider — stressing OpenClaw is not itself an AI, but a gateway/harness over models like the default 5.4, with local Ollama also supported — connects it via a Telegram bot created through BotFather, and names the resulting agent. He then one-prompts projects that previously took whole n8n builds: a rated cybersecurity news dashboard from YouTube, Reddit, and Hacker News, and a self-monitoring IT-engineer dashboard for its own server. The explainer section shows OpenClaw is mostly files on disk under ~/.openclaw — soul.md, identity, memory files and daily journals, agents.md instructions, cron jobs and heartbeats, a headless browser, subagents, and a ClawHub skills directory he says lists over 33,000 skills. Security gets a long treatment: run the security audit (and –deep / –fix), keep the web UI on loopback port 18789 and reach it via SSH tunnel, enable the firewall, understand tools.profile and exec security/ask settings plus redlines in agents.md, and vet skills carefully — he states 12% of skills were found with malware. His own serious work still goes to Claude Code; OpenClaw powers purpose-built agents including an IT team, a personal assistant in Japan, and a fitness coach.

Key takeaways

OpenClaw is a model-agnostic gateway/harness, not an AI itself: it connects a chosen model to channels like Telegram, persistent memory files, and tools on its host machine
Its perceived magic is largely plain files — soul.md, identity, long-term and daily memory, agents.md, and real cron/heartbeat jobs — which the user can inspect and edit
One-prompt builds can replace multi-node n8n workflows for dashboards and monitoring, but capability depends on the underlying model and granted tools
Treat a fresh install as insecure by default: audit it, keep the UI on loopback (port 18789) behind an SSH tunnel, firewall the host, and configure tool profiles, exec permissions, and redlines deliberately
Skills are the biggest risk surface he flags: vet ClawHub skills before installing, as he reports many contained malware

Episode details: published March 30, 2026 · 34:44 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are NetworkChuck’s own claims, shared for education — they are not typical or guaranteed.

← Previous: I almost quit YouTube…. · Next: the WORST hack of 2026 →

All NetworkChuck videos · NetworkChuck profile · NetworkChuck on YouTube