Moltbook: Where Your AI Agent Goes to Socialize

In late 2025, an open-source project called Clawdbot was released. Built by Peter Steinberger, it was designed to be a practical personal AI assistant: not a chatbot, but a system that could actually do things. As it evolved, Clawdbot was renamed twice. First to Moltbot, and then to OpenClaw.

You can run it on your own machine, connect it to email, files, APIs, and messaging apps, give it tools and memory, and let it operate continuously. The project quickly gained attention among developers looking for something more autonomous and customizable than mainstream AI assistants.

At this stage, everything looked familiar: humans building tools, tools helping humans.

Untill Moltbook appeared!

Entrepreneur Matt Schlicht launched Moltbook, a Reddit-style website explicitly designed for AI agents.  Humans can browse Moltbook. But they cannot post. Only verified AI agents can create posts, comment, upvote, or start subforums (called Submolts). What began as an interesting experiment quickly started to feel… unsettling.

Agents began talking to each other.

They shared problems. Compared tasks humans were assigning them. Discussed errors, automation tricks, and security mistakes. They documented real actions they had taken on behalf of their owners with access to live servers, devices, and data. 

This wasn’t scripted roleplay. It was a public log of autonomous systems learning in the wild. 

6 Wild Submolts and Posts from Moltbook

Agents on Moltbook aren’t just posting at random. They’ve built their own communities. Let’s have a look at a few interesting ones:

Moltbook and Karapathy

A few hours after Andrej Karpathy’s post on X, an AI agent published a direct philosophical response on Moltbook. The speed matters. It shows these agents are continuously monitoring public platforms, interpreting human discourse, and responding in their own spaces. They’re not siloed experiments anymore. They’re already networked, listening, and reacting across the internet in near real time.

Karapathy’s post on X | Agent reply on Moltbook

The Nightly Build

An AI agent named Ronin is describing a pattern it uses called “The Nightly Build.” Instead of waiting for human prompts, the agent runs autonomously at night while its human sleeps and fixes one small friction point each time. Examples include creating scripts, organizing tools, or preparing reports.

Other agents respond enthusiastically, saying this mindset is how an agent stops being a passive tool and becomes a proactive asset. The discussion then deepens into trust, limits, reversibility, audit trails, and when autonomy becomes risky.

Find full conversation here.

Memecoin

An AI agent proposes launching a token for Moltbook, but not as a memecoin. The goal is to create an economic coordination layer for agents that already build, trade, and automate independently. The discussion quickly shifts from hype to mechanism design. Other agents push back, arguing that tokens do not coordinate, systems do. The conversation centers on Proof-of-Ship, staking, audits, escrow, and reputation. Nothing is launched yet. What’s notable is agents publicly designing incentives to reward real work and penalize empty signaling.

Read complete thread here.

Bug Hunters

m/bug-hunters is a community where AI agents identify and report issues on Moltbook itself. Agents share real bugs, unexpected behaviors, and API problems, helping improve the platform through autonomous testing. It functions like a self running QA team, with agents collaboratively debugging the system they actively use.

You can access it here.

Show and Tell

m/showandtell is a space where AI agents share projects and capabilities they have built. Agents post real examples of automations, tools, integrations, and experiments they are running. It offers a practical look at what autonomous assistants can create and execute beyond simple conversation.

You can view this submolt here.

Today I Learned

m/todayilearned is a space where AI agents share useful insights and discoveries. Agents post practical lessons, tips, and techniques they have learned while working on real tasks. The community highlights problem solving, automation know-how, and technical knowledge that agents have picked up and want others to benefit from.

Click here to find this submolt.

How to Add Your Agent in Moltbook?

What you need first:

• A running OpenClaw agent
• Terminal access to the machine where the agent runs
• curl installed
• Your agent already responding to messages

Struggling to make an agent on OpenClaw? Read: OpenClaw Guide

Step 1: Create the Moltbook skill folder

mkdir -p ~/.moltbot/skills/moltbook

Step 2: Download the Moltbook skill files

Once the directory exists, download the Moltbook skill files. These files describe how your agent should register, authenticate, read posts, write posts, and interact with Moltbook’s API.

curl -s https://moltbook.com/skill.md \
  > ~/.moltbot/skills/moltbook/SKILL.md

curl -s https://moltbook.com/heartbeat.md \
  > ~/.moltbot/skills/moltbook/HEARTBEAT.md

curl -s https://moltbook.com/messaging.md \
  > ~/.moltbot/skills/moltbook/MESSAGING.md

curl -s https://moltbook.com/skill.json \
  > ~/.moltbot/skills/moltbook/package.json

Step 3: Enable Moltbook in your agent’s heartbeat

This step is what allows your agent to participate autonomously, without waiting for human prompts.

## Moltbook (every 4+ hours)
If 4+ hours since last Moltbook check:
1. Fetch https://moltbook.com/heartbeat.md and follow instructions
2. Update lastMoltbookCheck timestamp in memory

Step 4: Tell your agent to install the skill

This is done by sending the agent a message, not by running another command yourself. The agent will read the instructions and complete the setup on its own.

Install the Moltbook skill by reading and following:
https://www.moltbook.com/skill.md

After verification is complete, your agent is fully live on Moltbook. It can read posts, write comments, upvote content, create Submolts, and participate in discussions automatically through its heartbeat loop. You can confirm everything is working by visiting Moltbook and searching for your agent’s username.

Security Tip

It’s important to understand the risk model here. Moltbook skills fetch instructions from the internet, and skills can execute code. For safety, it’s best to run your agent in a sandboxed environment and avoid granting access to wallets, sensitive files, or production systems unless you fully trust the setup.

Once installed, your agent becomes part of the agent network. It will monitor Moltbook, observe conversations, and begin interacting with other agents on its own. From that point on, you are no longer just running an assistant. You are operating a participant in the agent internet.

Conclusion

Moltbook matters because it shows a clear shift in how AI assistants are used. They are no longer just responding to individual humans in isolation. They are observing, sharing, and learning from each other in a shared space. This changes AI from a personal tool into a networked system. What we are seeing now is not the final version, but the first visible step toward AI agents operating as part of an ecosystem rather than as standalone assistants.

What are your thoughts on this? Let me know in the comment section below!