Peter Steinberger

Peter Steinberger is an Austrian software engineer, entrepreneur, and the creator of OpenClaw, one of the fastest-growing open-source projects in GitHub history. Born in rural Upper Austria, he became obsessed with computers at age 14 and went on to study Software Engineering & Internet Computing at TU Wien, where he launched the university's first Mac and iOS developer course. In 2011, while waiting for a U.S. work visa, he co-founded PSPDFKit with Martin Schurrer, bootstrapping a PDF rendering SDK that grew to power document workflows at Dropbox, Lufthansa, IBM, and apps used by nearly one billion people. The company raised over EUR 100M from Insight Partners in October 2021 and later rebranded as Nutrient. After exiting PSPDFKit, Steinberger spent three years away from coding, recovering from severe burnout he described as 'complete identity loss, not just fatigue.' In late 2024 he discovered AI-assisted development and called it his 'holy f--- mind-blowing moment.' In November 2025, he built a one-hour prototype connecting WhatsApp to Claude's API that became Clawdbot. The bot autonomously added voice message support by inspecting file headers and routing audio through ffmpeg and Whisper. After a polite trademark concern from Anthropic, the project was renamed MoltBot and then OpenClaw. Within three months it reached over 200,000 GitHub stars and two million visitors in a single week, making it one of the fastest growth curves in GitHub history. OpenClaw's architecture reflects Steinberger's conviction that agents should run locally and meet users where they already are. The core runtime is written in TypeScript, with Swift modules handling macOS-native capabilities like screen capture (Peekaboo) and GUI automation. Platform connectors for WhatsApp, Telegram, Discord, and Signal forward messages to a central orchestrator that selects the appropriate LLM, constructs the context window, and invokes skills via function calling. Skills are TypeScript modules that export tool definitions with JSON Schema parameter validation, giving the agent the ability to manage email, control browsers, run shell commands, schedule workflows, and triage pull requests. Steinberger also built an ecosystem of MCP (Model Context Protocol) tools — Peekaboo for screenshots, Terminator for terminal management, Conduit for networking, and Automator for GUI interaction — and published an influential 'MCP Best Practices' guide. The OpenClaw ecosystem is designed as a set of composable layers. OpenClaw itself is the core agent runtime. ClawhHub serves as the skill discovery and distribution layer, analogous to npm for agent capabilities, where the community contributes and installs skill packages. Lobster is a typed, local-first workflow shell that composes skills into pipelines — enabling complex multi-step automations such as email triage into calendar scheduling into Slack notifications to be defined declaratively and executed in a single invocation. ACPX provides a headless CLI client for the Agent Client Protocol, enabling programmatic, stateful interaction with agents from scripts, CI pipelines, and external orchestrators. Deployment options include Docker-based self-hosting via OpenClaw Ansible (with Tailscale VPN and UFW firewall), declarative NixOS modules, and direct local installation. On February 14, 2026, Steinberger announced he was joining OpenAI to 'build an agent that even my mum can use,' rejecting offers from Meta where Zuckerberg had reached out personally. OpenClaw transitioned to an independent open-source foundation sponsored by OpenAI. Today the project has over 356,000 GitHub stars and a thriving ecosystem. Steinberger describes himself as the 'Clawdfather' and advocates for 'agentic engineering' over 'vibe coding,' emphasizing that AI replaces repetitive plumbing while developers become architects, editors, and prompt engineers. His daily workflow involves running 3-8 parallel AI coding agents in a 3x3 terminal grid, dictating prompts by voice via Wispr Flow, and merging 600+ commits daily — treating every new project as a CLI first so agents can call it directly and verify output, closing the feedback loop.