OpenClaw

Open-source personal AI agent created by Peter Steinberger in late 2025. Now one of the most-starred GitHub projects. Hub-and-spoke architecture centered on a Node.js Gateway, with adapters for Slack/iMessage/WhatsApp/etc., a tools layer, and a markdown-based skills system.

Why it appears in 5 sources in this wiki

OpenClaw is a hub entity. It's the canonical personal agent example across this wiki:

• LLM Wiki (Karpathy gist) — listed as a canonical agent for the LLM Wiki Pattern
• Andrej Karpathy on Agentic Engineering (Sequoia AI Ascent) — install-via-prompt example for Software 3.0
• Agentic AI in the Enterprise (Praveen Akkiraju, CXOTalk) — "the first open-source framework that actually worked" as a personal agent; "truly blew people's minds"
• Printing Press (Nate Herk video) — Steinberger built gog CLI as the founding insight that became Printing Press
• What is OpenClaw (IBM Technology) — primary source page; full architecture walkthrough

Architecture (per What is OpenClaw (IBM Technology))

• Gateway — central Node.js WebSocket service; routing, sessions, multi-agent management, tool dispatch
• Adapters — unify incoming sources (Slack, iMessage, Teams, Discord, WhatsApp, etc.) into one internal format
• Tools layer — built-in browser automation, terminal access
• Skills — markdown folders with workflow instructions; metadata loaded by default, full skill read on-demand (avoids context bloat — same insight as CLI vs API vs MCP)
• Config — agents.md and sole.md (analogous to Claude Code's CLAUDE.md)

Operational model

• Implements the Agentic Loop (ReAct: reason → act → observe → repeat)
• Runs locally on Node.js (laptop, VM, Raspberry Pi)
• Can be triggered ad-hoc via messaging or via cron-like automated jobs

Security notes

• Local execution + filesystem + terminal + integrations = potential backdoor if misconfigured
• Thousands of internet-exposed instances exist due to misconfig
• Prompt injection is the dominant threat vector
• Recommendations: isolated environments, audit skills, encrypt credentials before sending to LLM

2026-08-02 — retrospective read: the mass-learning inflection point

Nathaniel Whittemore in 6 Questions Shaping Enterprise AI (AI Daily Brief) makes the historical case that OpenClaw's significance was pedagogical rather than product:

*"Hundreds of thousands of people, perhaps millions if you include the people who are standing in line in China to get access to an Open Claw, really got their hands dirty figuring out the guts of how these agents work. And while you don't necessarily see everyone running their Mac Mini setups anymore, the explosive learning of that early period of Open Claw, I think will be seen as a key inflection point moment for the history of agentic AI."*

The claim: OpenClaw is where a deepened, practical understanding of what an agent actually is — and what it means to build and manage one — spread beyond the labs. Whittemore pairs it with Claude Code's 2025 growth and OpenAI's Codex push as the harness-literacy wave that preceded the enterprise agentic turn. The declining visibility of self-hosted setups is, on this read, not a failure signal but the normal arc of a teaching technology.

He also uses it as the era's unit of scale: last year's headline stat was Google processing ~a quadrillion tokens; now "a quadrillion tokens at this point is about what a single open claw left unattended will do in a month."

Sources

• What is OpenClaw (IBM Technology) (primary)
• Andrej Karpathy on Agentic Engineering (Sequoia AI Ascent)
• Agentic AI in the Enterprise (Praveen Akkiraju, CXOTalk)
• LLM Wiki (Karpathy gist)
• Printing Press (Nate Herk video)
• 6 Questions Shaping Enterprise AI (AI Daily Brief) — 2026-08-02; the "mass-learning inflection point" retrospective