Overview
- A new five‑part field guide explains how to build a multilingual WhatsApp agent on OpenClaw, covering model setup, a loopback‑only gateway, Docker deployment, session handling, and deny‑first tool permissions.
- Developers shipped GuardianClaw, a human‑in‑the‑loop email assistant that drafts replies but never sends without approval, with defenses like localhost binding, long tokens, Tailscale access, Docker sandboxing, and local Ollama models for privacy.
- ClawNet, a Windows security terminal, uses OpenClaw to watch live network connections, tie them to processes, classify risks with reasons, remember past threats, and push Telegram alerts with clear actions like killing a process or blocking an IP.
- A step‑by‑step Cloud9 and EC2 guide shows OpenClaw running with Ollama and a small LLaMA model on a modest CPU instance, keeping data on a user‑controlled VM and avoiding third‑party telemetry.
- Authors split on hardware needs, with one urging modern GPUs for snappy local agents and another reporting solid results on CPU, as the project’s creator frames the shift from chat to goal‑seeking agents that plan, retry, and act across real tools.