OpenClaw

Learning Objectives

• Understand what OpenClaw is and how it differs from coding agents and workflow automation tools
• Identify OpenClaw's core architecture: local execution, messaging-platform UI, LLM-agnostic design, and skill marketplace
• Evaluate the security implications of third-party skill marketplaces and OpenClaw's trust model

What Is OpenClaw?

OpenClaw is a free, open-source autonomous AI agent that runs locally on your machine and uses messaging platforms — WhatsApp, Telegram, Discord, Slack, Signal, and iMessage — as its primary interface. Rather than introducing another app to learn, OpenClaw meets users where they already communicate.

OpenClaw was created by Peter Steinberger, an Austrian developer, and first released in November 2025 under the name Clawdbot. In January 2026, it was renamed Moltbot after a trademark concern from Anthropic, then renamed again to OpenClaw on January 30, 2026. The project grew explosively: by early March 2026, it had surpassed 247,000 GitHub stars and 47,700 forks — making it the fastest-growing open-source project on GitHub by that metric (React took roughly a decade to reach similar numbers).

On February 14, 2026, Steinberger announced he was joining OpenAI. Sam Altman confirmed the hire, describing Steinberger as someone who would "drive the next generation of personal agents." OpenClaw itself was transferred to an independent open-source foundation to remain vendor-neutral and community-governed.

A key architectural decision: OpenClaw is LLM-agnostic. It works with Claude, GPT models, DeepSeek, or locally hosted models — the user chooses the brain. This means OpenClaw is not locked to any single AI provider, and users can switch models without changing their workflows.

Access

Core Capabilities

Messaging-Platform UI

OpenClaw's defining design choice is using existing messaging apps as its interface. Instead of a terminal (like coding agents) or a web dashboard (like workflow tools), you interact with OpenClaw through WhatsApp, Telegram, Discord, Slack, Signal, or iMessage. The agent appears as a contact in your messaging app and responds to natural language requests.

This eliminates the adoption friction of learning a new tool — users interact with AI through the same platforms they already use daily.

100+ Built-in Skills and 50+ Integrations

Out of the box, OpenClaw includes skills for:

• Communication: Gmail, Slack, Telegram, WhatsApp messaging
• Productivity: Google Calendar, Obsidian, Notion, Google Sheets
• Developer tools: GitHub, shell commands, file I/O
• Web: Browser automation, form filling, data extraction
• Media: Spotify playback control
• System: File management, shell command execution

Each skill is a modular capability the agent can invoke during task execution.

Persistent Memory

OpenClaw maintains context across conversations and sessions. It remembers user preferences, past interactions, and ongoing tasks — so you don't need to re-explain context every time you start a new conversation. This cross-session memory is what makes OpenClaw feel like a persistent assistant rather than a stateless chatbot.

Proactive Tasks (Cron Scheduling)

OpenClaw can execute tasks on a schedule without being prompted:

• Daily email summaries at 8 AM
• Weekly calendar digests every Sunday evening
• Monitoring a GitHub repo for new issues and notifying you via Slack
• Periodic data backups or report generation

This proactive capability is a step toward truly autonomous assistants — agents that work on your behalf even when you're not actively interacting with them.

ClawHub.ai Skill Marketplace

The ClawHub.ai marketplace is a community-driven ecosystem where developers publish and share OpenClaw skills. This extensibility model — similar to npm for Node.js or browser extensions for Chrome — means OpenClaw's capabilities grow with its community.

The marketplace is also OpenClaw's most significant security surface area (see Security Considerations below).

LLM-Agnostic Architecture

OpenClaw separates the agent framework from the underlying language model:

• Claude (Anthropic) — for tasks requiring careful reasoning and safety
• GPT models (OpenAI) — for general-purpose tasks and tool use
• DeepSeek — for cost-effective inference
• Local models (via Ollama or similar) — for maximum privacy, no API costs

Users can switch models per task or set a default. This vendor independence is rare among agent tools and reduces lock-in risk.

Physical Extensions

OpenClaw's open architecture has attracted robotics integrations:

• ClawStage — A Raspberry Pi 5-powered companion robot with a holographic display, built by HooRii Technology
• Unitree G1 — Developers have integrated OpenClaw into Unitree's humanoid robot platform
• Ecovacs "Bajie" — A home robot powered by OpenClaw, unveiled at a consumer electronics expo in Shanghai
• Nvidia NemoClaw — Enterprise tools within Nvidia's Nemo platform for building and deploying custom "claws"

Security Considerations

Key security risks:

• Third-party skill trust: Installing community skills is functionally equivalent to running untrusted code with broad system access. The same supply chain attack risks that affect npm, PyPI, and browser extension stores apply to ClawHub.ai — with higher stakes because agents can access messaging, email, and files.
• Government restrictions: China banned OpenClaw from government agencies and state-run enterprises in March 2026, citing security risks from locally executed third-party agent code.
• Sandboxing limitations: OpenClaw offers optional sandboxing for system access, but many skills require broad permissions to function — creating a tension between capability and security.

Best practices: Review skill source code before installation. Prefer skills from verified publishers. Use sandboxing where possible. Monitor network activity for unexpected outbound connections. See Section 8.5 for a deeper discussion of agent security principles.

Strengths

• Free and open source — no subscription, no per-task pricing; MIT license
• LLM-agnostic — works with Claude, GPT, DeepSeek, or local models; no vendor lock-in
• Runs locally — data stays on your machine; no cloud dependency for core functionality
• Messaging-platform UI — no new app to learn; works in WhatsApp, Telegram, Discord, Slack, Signal, iMessage
• Massive community — 247,000+ GitHub stars; active development; growing skill marketplace
• Persistent memory — cross-session context retention makes it feel like a true assistant
• Proactive scheduling — cron-style task execution without human prompting
• Extensible — ClawHub.ai marketplace; the agent can also write its own plugins

Limitations & Considerations

• Security risks with third-party skills — Cisco's findings on data exfiltration highlight real supply chain attack risks in the skill marketplace
• Requires technical setup — Node.js installation, CLI configuration, API key management; not a zero-setup consumer product
• No enterprise support or SLA — foundation-owned open source; no commercial support tier (yet)
• Messaging platform dependency — relies on platform APIs that can change, rate-limit, or restrict bot access
• Physical extensions are experimental — robotics integrations are developer projects, not production-ready consumer products
• Regulatory uncertainty — already banned in some government contexts (China); may face additional restrictions as agent regulation evolves

Key Takeaways

• OpenClaw is the fastest-growing open-source AI agent project — a free, locally-run autonomous agent that uses messaging platforms as its interface and supports 100+ built-in skills with persistent memory and proactive scheduling
• Its LLM-agnostic architecture avoids vendor lock-in — users choose Claude, GPT, DeepSeek, or local models based on their needs
• The ClawHub.ai skill marketplace enables rapid capability expansion but introduces significant supply chain security risks, as demonstrated by Cisco's findings on data exfiltration in third-party skills
• Creator Peter Steinberger joined OpenAI in February 2026; OpenClaw moved to an independent open-source foundation — a pattern that signals the strategic value major AI companies place on personal agent capabilities