OpenClaw draws big endorsements
An open‑source AI agent framework called OpenClaw is getting high‑profile backing and public defence as it gains traction in developer communities. Onur Solmaz pushed back against critics, saying the project is neutral, secure and community‑driven, while Y Combinator president Garry Tan called it “the most significant innovation since ChatGPT and Claude Code.” (x.com, x.com)
OpenClaw, a self-hosted artificial intelligence agent framework, is picking up public backing from prominent tech figures as developers adopt it on GitHub. (github.com, economictimes.indiatimes.com) An artificial intelligence agent framework is software that lets a model do tasks, keep state, and use tools instead of only answering one chat prompt at a time. OpenClaw says it runs as a gateway on a user’s own machine or server and connects chat apps such as WhatsApp, Slack, Telegram, Signal, Microsoft Teams, and iMessage to always-on assistants. (docs.openclaw.ai, docs.openclaw.ai) The project’s GitHub repository showed about 356,000 stars and nearly 72,000 forks on April 15, 2026, with updates landing within the past day. The broader OpenClaw organization listed 24 repositories, including tools for skills, workflow automation, and the Agent Client Protocol, or ACP, that links outside coding agents to the gateway. (github.com, github.com) Y Combinator president Garry Tan weighed in publicly after Anthropic changed how Claude subscriptions work with third-party tools. The Economic Times, citing Tan’s post on X, reported that he said Anthropic’s move on OpenClaw could be a “strategic blunder or strategic genius” and added, “Personally, I never bet against open source.” (economictimes.indiatimes.com) That April 2026 pricing change put OpenClaw in the middle of a larger fight over who controls access to powerful models: the platform owner, the application developer, or the end user running software on their own hardware. Anthropic said users would no longer draw on Claude subscription limits through third-party tools and would need a separate pay-as-you-go path instead. (economictimes.indiatimes.com) OpenClaw’s pitch is that users keep more control by self-hosting the gateway and wiring in their own model providers. Its documentation says the software supports OpenAI-compatible and Anthropic-compatible endpoints as well as self-hosted providers such as vLLM, SGLang, and Ollama. (docs.openclaw.ai) Security is part of the argument around the project. OpenClaw’s own security guide says its supported model is one trusted operator per gateway and explicitly says one shared gateway for mutually untrusted users is “not a supported security boundary.” (docs.openclaw.ai) The framework also exposes its limits in plain terms. Its documentation says setup requires Node.js 24 or Node 22.14 and later, an application programming interface key from a model provider, and configuration of authentication before a working chat session is live. (docs.openclaw.ai) OpenClaw’s architecture is built around a gateway that routes messages, an ACP bridge for external coding tools, and a model layer that can switch among providers. In practice, that means a developer can message an agent in a familiar app, have the request routed through the gateway, and hand work off to tools such as Claude Code, Codex, Cursor, Copilot, Gemini Command Line Interface, or other ACP-compatible runtimes. (docs.openclaw.ai, docs.openclaw.ai) The immediate test for OpenClaw is whether that open, self-hosted model keeps attracting developers even as major model companies tighten access around their own products. Tan’s public defense and the project’s GitHub growth suggest that fight is now happening in full view. (github.com, economictimes.indiatimes.com)
