Vending machine run by OpenClaw agent
An OpenClaw AI agent was shown autonomously operating a San Francisco vending machine—setting prices, selecting products, driving ads and reporting sales to a live dashboard. The demo is an example of agents making real‑world, revenue‑driving decisions in a narrow physical workflow. (x.com)
A San Francisco vending machine was shown running under an OpenClaw agent that picked products, set prices, generated ads and logged sales to a live dashboard. (24vids.com) The public video, posted by Om Patel and viewed more than 308,000 times in roughly 11 hours, describes an “actual physical vending machine” operated by the agent. The clip says OpenClaw decided what to sell, how to name items, how to price them, and how to track revenue. (24vids.com) OpenClaw is open-source software that runs as a self-hosted gateway on a user’s own machine or server and connects chat apps like WhatsApp, Slack, Telegram and iMessage to an always-available agent. Its documentation says the system includes a browser dashboard for chat, configuration, sessions and connected nodes. (docs.openclaw.ai) That matters in a vending machine because the job is narrow and repetitive: choose stock, adjust pricing, promote items and count sales. The demo turns those software actions into a physical retail loop with a machine in one location and a dashboard that can be monitored remotely. (24vids.com) The project behind the agent has grown fast. The main OpenClaw GitHub repository showed about 356,000 stars and more than 30,000 commits when checked on April 15, 2026, with Peter Steinberger listed on GitHub as the founder behind OpenClaw. (github.com 1) (github.com 2) OpenClaw’s pitch is that an agent has “eyes and hands” on a computer, meaning it can use tools, maintain sessions and keep working across channels instead of only answering prompts. A vending machine is a simpler version of that idea: the agent is not running a whole store, but it is making operating decisions inside one fixed workflow. (openclaw.ai) (docs.openclaw.ai) The tradeoff is control and security. GitHub’s advisory database lists a long run of OpenClaw vulnerabilities disclosed in 2026, including issues tied to server-side request forgery, browser-origin bypasses and token exfiltration, with patched versions released across March and April. (github.com 1) (github.com 2) (github.com 3) The vending machine demo does not show those risks being exploited, and the advisories repeatedly note that OpenClaw is designed as a user-controlled local assistant rather than a multi-tenant cloud service. But the clip does show what agents look like when they move from inbox tasks to a cash register, one snack at a time. (github.com) (24vids.com)
