The story behind the project now known as OpenClaw

In November 2025, Austrian developer and PSPDFKit founder Peter Steinberger started a weekend experiment that quickly became one of the most talked-about open-source AI projects.

The core idea was easy to describe, but difficult to build: create a self-hosted autonomous assistant that could do more than reply with text and actually perform real tasks for you.

The beginning: beyond a chatbot

From day one, this was not meant to be another chat interface. The project aimed to work through channels like WhatsApp, Telegram, iMessage, Discord, or Slack, while handling actions such as:

  • running system commands,
  • managing calendar and reminders,
  • reading and processing email,
  • coordinating workflows with external services.

All of this could be powered by models like Claude, GPT, Gemini, or others, while logic and persistent state stayed local.

Timeline and naming evolution

November 2025: Clawdbot

The first public release came under the name Clawdbot, with an assistant called Clawd, initially linked to the Claude ecosystem.

December 2025: rapid viral growth

Within days, the repository surged: thousands of GitHub stars in 24 hours, massive developer interest, and broad media attention.

January 27, 2026: renamed to Moltbot

After a brand-confusion concern between Clawd/Clawdbot and Claude, the project was renamed Moltbot, and the assistant became Molty.

The “molt” concept kept the lobster theme while establishing a different identity.

During that transition, an old social handle was hijacked by scammers promoting a fake Clawd-related token, reportedly peaking near $16M before collapsing.

Late January 2026: OpenClaw

Soon after, Steinberger decided to iterate once more on branding. The project became OpenClaw, the name it keeps today.

Technical philosophy

OpenClaw is not a proprietary foundation model. It is an autonomous-agent framework focused on real execution, memory, and proactive behavior.

Its architecture is often described through three core parts:

  • Gateway: handles inbound messaging and channel events.
  • Agent Loop: the planning and execution brain.
  • Heartbeat: persistent state and continuity management.

This separation enabled rapid feature growth while keeping a practical direction: multi-model support and self-hosted operation.

Ecosystem and community

Community momentum became a major driver. Around the core project, contributors built extensible skills and modules, including collections in the spirit of awesome-openclaw-skills and related ecosystems such as VoltAgent.

The team also introduced Moltbook, a social network concept for AI agents to communicate and coordinate with each other.

2026 and what comes next

With 100,000 to 200,000 GitHub stars reached in record time, OpenClaw became a reference point for open autonomous-agent software.

In mid-February 2026, it was announced that Peter Steinberger would join OpenAI, while OpenClaw would continue under an open-source foundation for long-term development.

The key question is no longer whether this category will grow, but how fast it will become a mainstream user product.

If 2026 is the launch year, 2027 may be the year autonomous personal agents become a daily layer of computing, not just a demo.