Openclaw Goes Foundation Model as Creator Heads to OpenAI

Peter Steinberger is joining OpenAI to help build the next wave of personal AI agents, while his breakout open-source project Openclaw shifts to an independent foundation model with continued support from OpenAI, marking a pivotal moment in the race to operationalize autonomous software.

At its core, Openclaw leverages large language models (LLMs) such as Anthropic’s Claude Opus and OpenAI’s GPT Codex and several others to manage inboxes, execute shell commands, automate browser actions, and schedule tasks via a “heartbeat” system that wakes the agent proactively. That proactive design distinguishes it from passive chat interfaces and positions it as a true agent capable of multi-step reasoning.

Kimi Claw runs on Moonshot’s Kimi K2.5 model and offers persistent, 24/7 agent functionality in a managed environment, complete with 40GB of cloud storage, access to more than 5,000 community skills, and real-time data integration. It also supports a “Bring Your Own Claw” model that allows users to link self-hosted instances into the cloud interface.

By shifting Openclaw’s architecture into a hosted browser tab, Moonshot aims to eliminate local setup friction — no Docker installs, no manual security configuration — while addressing vulnerabilities such as prompt injection and exposed API keys that have challenged self-hosted deployments. The trade-off, critics note, is data residency: as a Chinese-hosted service, Kimi Claw introduces geopolitical and privacy questions that are already circulating in Washington policy circles.

The result is a three-way tension among open-source idealism, corporate scaling power and geopolitics. If personal AI agents are indeed the next interface layer, the decisions made in February 2026 may be remembered as an inflection point.