How a Retired AI Veteran Built the Viral Moltbot Just for Fun

Peter Steinberger, the creator of PSPDFKit, retired after selling his company and later rediscovered his passion for AI. He built Moltbot (originally Clawdbot) purely for enjoyment, without funding or a commercial plan, and it quickly amassed over 32 k stars on GitHub, prompting even non‑technical users to buy Mac Minis to run it.

From Burnout to Addiction: Reclaiming the Mojo

Four years after selling his company, Steinberger experienced severe burnout. In April 2025 he tried early AI tools like Claude Code’s beta, which reignited his excitement and led to an "addiction" to building AI‑powered utilities for fun, such as creating the "world's most expensive alarm clock" that remotely wakes him via a London‑hosted AI agent controlling his MacBook.

Technical Philosophy: CLI Is the Agent’s Native Language

Steinberger argues that traditional GUIs cater to human bandwidth limits, while AI agents can read manuals and remember parameters. Consequently, Moltbot adopts a radical CLI‑First strategy: every capability is exposed as an atomic command‑line tool with a --help interface, allowing agents to learn and invoke them instantly.

Control a Sonos speaker via a CLI.

View home cameras with a CLI.

Query Google Maps with a CLI.

The agent composes these CLIs to act in both digital and physical worlds, offering efficiency far beyond mouse‑based RPA.

Breaking the Walls: Data Liberation

Moltbot challenges big‑tech walled gardens by using AI to bypass closed APIs. For example, Steinberger hacks WhatsApp’s desktop protocol so that when a voice message arrives, Moltbot automatically downloads it, converts it with ffmpeg, transcribes via Whisper, extracts restaurant details with OpenAI, and adds the location to Google Maps—all without user intervention.

He asserts that "Apps will eventually melt away" as AI agents consume their functionality via APIs or hacks.

Local‑First: Privacy and Performance Trade‑offs

Although he leverages OpenAI and Anthropic models, Steinberger invests heavily in local models (e.g., minimax 2.1) and equips his Mac Studio with 512 GB RAM. He emphasizes that running agents locally eliminates compliance hurdles and keeps data and models under the user’s control.

Data resides on your hard drive.

Models run on your GPU.

Operations execute within your OS.

He predicts that widespread AI agents will boost high‑performance hardware sales, describing it as "the liberation of data."

Business vs. Open Source: Building for Love, Not Profit

Following Moltbot’s popularity, venture capitalists and large companies approached Steinberger, but he maintains a non‑profit stance: "I built this for me." He believes code itself holds little monetary value; instead, ideas, community, and brand are what matter.

Steinberger envisions Moltbot operating as a foundation—a community‑driven playground rather than a proprietary ecosystem.

Conclusion: Build Your Own Agentic Loop

Steinberger urges developers to stop merely observing and start constructing their own agentic loops: write CLI tools, hack APIs, and equip AI with memory, tool use, and autonomy. In this era, imagination, not technical barriers, limits what can be achieved.