Two Free Agents: One Lives in Chat Apps, the Other Writes Skills as It Learns

datawebot posted a tweet listing two Agents. At the start, the tweet notes that some Agents live in the cloud, and some live on your laptop. Here, “Agent” can be understood as an AI program that can call tools and execute tasks on its own.

OpenClaw, about 393k GitHub stars. It lives in WhatsApp, iMessage, Slack, and 20+ other entry points. Hermes, from Nous Research, about 256k stars, writes its own skills during learning. Both are free and MIT-licensed.

The tweet ends with one piece of advice for peers: sandbox first.

That sentence is more worth reading than the two numbers above. When an Agent is resident in chat software, its entry point is the dialog box people already use every day. When it can write its own skills, its capability boundary is not fixed. Free and MIT lower the barrier to trying it, and also push permissions, isolation, and sandboxing to the fore. GitHub stars only indicate attention; they cannot serve as a quality guarantee. If you’re really going to run it, set up the sandbox first.

The tweet includes a vertical video whose cover text reads "Muse and Dots run on big tech's computers." On one side are big companies’ machines; on the other are chat entry points and the local environment—the contrast is quite clear.

For developers, the MIT license and star counts are signals, but don’t treat star counts as a benchmark. For ordinary users, it can be understood as two free, MIT-licensed assistants: one follows chat software, and the other extends its own skills while learning. If you want to try them, don’t give full permissions first. Sandboxing is not optional.

Original tweet: datawebot’s tweet