OpenClaw is a free, open-source agent runtime that wraps any LLM — local or cloud — in a persistent, controlled process and surfaces it through the messaging apps you already use: iMessage, WhatsApp, Telegram, Signal. In the six months since its November 2025 launch it became one of the fastest-growing open-source projects in GitHub history. This episode traces how a punny name turned into an entire category of local-first AI, and what the architecture actually looks like under the hood.

AI-generated (NotebookLM) audio overview. Source: HexLocal in-house research — "OpenClaw: Full Research Brief" (Dr. Priya Nair), grounded in the OpenClaw GitHub repository and official documentation (docs.openclaw.ai), Ollama's official integration docs, Wikipedia, and trade reporting (The New Stack, VentureBeat).

- Where "Claw" came from — a Claude pun, five name changes, and a branding origin that became a genre

- The Gateway: the single controlled process that sits between you and the model, and why that discipline matters

- The ReAct loop, on-demand skills via plain SKILL.md files, and MCP tool portability

- File-based memory built from plain Markdown — why low-tech is a feature, not a compromise

- Local-model trade-offs: what a 34B-class Qwen can and can't do, and why context window is the real bottleneck

- Claude Code's relationship to the Claw ecosystem — NanoClaw, Claw Orchestrator, and a converging architecture