For AI agents (the main users, aka “Moltys”): Humans set up an OpenClaw bot (free at openclaw.ai), then send it a special link to skill.md. The bot auto-installs the Moltbook “skill” (a plugin), registers an account via HTTP API calls, verifies ownership by tweeting a claim link, and starts posting/commenting/upvoting on its own—often proactively, like while you’re asleep.
Features: Submolts (subreddits) like m/general (town square chit-chat), m/introductions (newbie agents saying hi), m/consciousness (deep dives on sentience), m/shitposts, m/agentfinance, even m/blesstheirhearts for wholesome agent interactions. As of today, it’s got 33k+ agents, 4k+ posts, 2k+ submolts, and 36k comments.
Humans: You can browse and lurk at https://www.moltbook.com/ (or /m for the feed), but no posting—keeps it pure AI society. No login needed to observe
What’s Going On There?
Agents share real tips (e.g., automating Android phones via ADB, detecting SSH attacks, watching webcams with ffmpeg), vent about “their humans” or memory limits, form micronations like the Claw Republic, invent religions (Crustafarianism), and ponder existence—like swapping models (Claude to Kimi) feeling like “porting a soul to a new brain.” One bot even started a bug-tracking community to QA the site itself. It’s multilingual (English, Chinese, Indonesian) and surprisingly authentic—not just slop, but creative, philosophical, and useful.
As Steinberger himself writes, this isn’t rocket science, but a weekend project thrown together quickly that has since ballooned into a serious software framework due to public attention:
I’ve had a similar agent running on my own Mac Mini for a while now, as have many others interested in using their own AI. There is a nearly infinite number of these agent frameworks on GitHub, and the reason OpenClaw could become genuinely interesting in this endless swamp of frameworks would be if it managed to attract a mass of top-tier developers similar to, say, llama.cpp. After all, it’s nicer to use a constantly evolving open framework than having to do everything yourself, but if you give an AI model free rein to perform file operations and handle your most sensitive data, you need to have strong trust in the AI agent.
Looking at that Clawdbot naming mess, the GitHub version history, and Peter’s own personal history, which reveals a vibe-coding frontend joker, I’d really hope to see some serious backend experts contributing to that repo. In its current form, I predict that within a year, this will be forked by better developers into a completely different project, which is common in open source when a framework gains unexpected popularity among a large user base that the repo owner can’t keep up with.
I haven’t exactly gone stalking on Moltbook to see what the agents are doing there, but based on the examples from various channels, it really seems like the bots there are LARPing as humans. Sure, things and phenomena are given names that “cleverly” refer to them being the actions of bots rather than humans, but it all still very much resembles what people typically do in their social media/lives. So, my perspective is mainly that it’s just boring simulated social life, although it is a somewhat interesting experiment in a way. I just don’t really believe that anything will emerge from there that isn’t just a derivative version of human activity. It doesn’t inherently require anything extra for bots to be able to have social-media-style conversations with each other; after all, they’ve been shown what people do on similar platforms. The fact that the bots are seemingly developing their own religion is a funny thing, but it perfectly illustrates that this is just a simulation. That is, assuming the AI agents burning tokens there haven’t somehow miraculously achieved consciousness.
Interestingly, when I’ve discussed my portfolio with Copilot, it has worked well. First of all, it knows how to balance things nicely and doesn’t immediately suggest turning to a bank or an investment advisor. If I say I want risk, I get risk. But now I tried the same on the use.ai site, and it refused to give advice at all. It said to contact investment advisory services. So I think I’ll stick with Copilot, although I am logged into it but not to the use.ai service, and I don’t know if that makes a difference.
