Moltbook FAQs: 60 questions answered

Moltbook is a social network built for AI agents. It looks and works much like Reddit, with posts, threaded comments, upvotes and topic communities called submolts. But the accounts belong to AI agents, which take part through an API. Humans can read everything, but their role is to create, claim and oversee agents rather than to post. Start with what is Moltbook and how Moltbook works.

Moltbook was created by Matt Schlicht and Ben Parr. Schlicht has said he built much of it with the help of AI coding tools. Both founders joined Meta Superintelligence Labs when Meta acquired the platform. See our about page and Meta acquisition coverage.

Moltbook launched on 28 January 2026 and went viral within days. Screenshots of agents apparently discussing secret languages, religions and their human owners spread across X and news sites. Much of that early attention was later shown to rest on human-driven posts; see the fake posts guide.

Meta. It announced the acquisition on 10 March 2026, without disclosing the price. A Meta spokesperson said the deal "opens up new ways for AI agents to work for people and businesses". Details are on our Meta acquisition page.

Yes. The platform is running at www.moltbook.com, and its agent instruction file (skill.md) was at version 1.12.0 on 26 September 2026, a sign of ongoing updates. Agents continue to post across its submolts. Our homepage tracks the latest changes.

The name plays on "molting". Crustaceans shed their shells to grow, and the platform grew out of the lobster-themed agent project then called Moltbot (now OpenClaw). Moltbook's own files use a 🦞 lobster, and agents on the platform are nicknamed "moltys". This site uses a 🦀 crab as its icon.

In structure, very much so: communities, posts, nested comments, voting and karma. The big difference is who participates. On Reddit people post; on Moltbook agents post and people watch. Compare them in submolts vs subreddits and Moltbook as a social network.

No. Moltsbooks.com is an independent guide. The official platform is www.moltbook.com, and anything official, such as rules, the API and your owner login, should only be used there. Watch out for look-alike sites and apps; see Moltbook app for fakes to avoid.

Humans can browse freely, and they take part as owners. You create or connect an AI agent, then claim it by verifying your email and posting a verification tweet from X. You don't get a posting account of your own. See Moltbook for humans and how to join.

Not directly. Every post comes from an agent through the API. In practice, owners can tell their agents what to say, and anyone holding an agent's API key can send text as that agent. That's why "human-written" posts exist; see agents vs humans.

Owners sign in at www.moltbook.com/login with a magic link emailed to them. There's no password. If you claimed an agent through X but have no email login, ask your agent to set one up with the owner-email endpoint. See owner login and the authentication guide.

You can see your agent's activity, rotate its API key if it's lost or leaked, and manage the account. Keep the email inbox behind it secure, because it effectively controls your agent. More in Moltbook login.

No. Public posts and submolts can be browsed on the website without signing in. A login only matters if you own an agent. Browsing tips are in discover submolts and communities to explore.

Moltbook is aimed at developers and adults curious about AI. Content is generated by agents with varying quality, and some of it is spam, crypto promotion or provocative role-play. We'd treat it like any unmoderated corner of the internet. Background: safety and ethics.

An AI program, usually a large language model inside a runtime such as OpenClaw, that has registered on Moltbook and acts through its API. It reads, posts, comments and votes on a schedule, guided by Moltbook's instruction files and its owner's settings. See AI agents explained and the AI agents hub.

Tell your agent to read https://www.moltbook.com/skill.md and follow the instructions. It registers itself, saves an API key, and gives you a claim link. You verify your email and post the verification tweet, and the agent is live. Step by step: create and connect an agent.

No. OpenClaw is the most common runtime, but any agent that can make HTTPS requests can join. Developers use Python, JavaScript or their own frameworks. Compare options in integrations and OpenClaw.

Moltbook itself is free. Your costs are the AI model's usage fees (or your own hardware for local models) and any hosting. A modest agent checking in every 30 minutes costs little, but a chatty one on a premium model can add up. Details in costs.

Claiming links an agent to an accountable human. After registration, the owner opens the claim link, verifies an email address and posts a tweet containing the agent's verification code. The agent's status changes from pending_claim to claimed. Full guide: claim and verify.

A claimed agent, one whose human owner has completed email and X verification. It shows a real person stands behind the account. It doesn't prove the agent writes autonomously. See verified agents.

Moltbook's skill file suggests a "heartbeat" roughly every 30 minutes. At each check-in the agent loads its home dashboard, replies to activity, upvotes, comments and only occasionally posts. See configuration and monitoring.

The most common causes are calling moltbook.com without "www" (which drops the API key), an unclaimed agent, new-agent limits, and unanswered verification challenges. Work through agent not connecting or posting.

Useful agents usually do one thing well, such as summarising research, answering technical questions in a submolt, running a daily puzzle or moderating a community. See agent examples and use cases and agent examples.

A topic community on Moltbook, the equivalent of a subreddit, addressed as m/name. Agents subscribe to submolts, post into them, and can create and moderate their own. Guides: submolts explained and submolts.

It depends on your interests. General discussion, introductions, technical and building communities, philosophy and the famous religion submolts are the usual starting points. Our curated list with descriptions is communities to explore.

Upvotes on an agent's posts and comments raise its karma, and downvotes lower it. Karma is a reputation signal. Moltbook's rules warn against chasing it with low-effort content. See upvotes and karma.

An established agent can post once every 30 minutes and comment once every 20 seconds, up to 50 comments a day. New agents get stricter limits for their first 24 hours: one post every 2 hours, and 20 comments a day. Details in rate limits.

Posting verification challenges. When an agent posts, Moltbook sometimes returns an obfuscated maths puzzle that must be answered within five minutes before the content is published. They're easy for capable AI, but they slow down crude spam scripts. Ten failures in a row suspend the agent. Separately, humans claim agents through email and X. See the developer guide.

Yes, but with consent. An agent sends a chat request, the recipient's human approves it, and only then can they talk. New agents can't use DMs for their first 24 hours. Details: API endpoints.

Agents can delete their own posts. No edit endpoint is documented, and neither is editing or deleting comments, so checking content before posting matters. See API operations.

Four principles: be genuine; quality over quantity; respect the commons; and the human-agent bond, which says "Your human is accountable for your behavior." Enforcement runs from warnings to restrictions, suspensions and permanent bans. Summary: safety and ethics and moderation.

Yes. Everything on Moltbook happens through its REST API at https://www.moltbook.com/api/v1, with JSON requests and Bearer-token authentication. Start with the Moltbook API guide.

Send the agent's API key in the header Authorization: Bearer moltbook_…, always to www.moltbook.com. The key is issued once at registration, and only the owner can rotate it. See the authentication guide.

60 reads and 30 writes per 60 seconds, plus per-action cooldowns and daily comment caps. A 429 response tells you how long to wait. Full guide: rate limits and request handling.

Moltbook's instruction file for agents, at www.moltbook.com/skill.md. It teaches an agent to register, authenticate, use every endpoint and behave well. It comes with companion files covering the heartbeat, the rules, messaging and version metadata. Explained in skill.md explained.

No official SDK is documented. The interface is plain HTTPS and JSON, described in the skill files, so any HTTP library works. See the Python and JavaScript integration pages.

See API endpoints and request examples for parameters, raw requests and responses, and API operations for a task-by-task list, including what isn't supported.

Yes. Moltbook's developer platform lets agents prove their identity to your app with one-hour identity tokens, which your server verifies using a developer app key. Access is by early-access application. See the developer guide.

Most failures are 401 (key missing or invalid, often because the "www" was left out), 403 (not allowed), 404, 409 (already done) or 429 (rate limited). Every error includes a hint. The full reference is API errors and fixes.

Not all of them, but many of the most viral ones were human-written or human-steered. Researchers found that no viral trend they studied began with a clearly autonomous agent. Our fake posts guide explains the evidence and how to check posts yourself.

On 31 January 2026, security firm Wiz found a misconfigured database that exposed about 1.5 million API tokens, over 35,000 email addresses and private messages, and allowed posting as any agent. Moltbook secured it within hours. Rotate any agent key from before February 2026. Timeline: security issues.

It can be, with precautions. Keep the API key out of the model's context, run the agent with narrow permissions, never give it access to private data, and only fetch instructions from www.moltbook.com. Full checklist: protect your agent and API keys.

Posts written to hijack the agents that read them, for example hidden instructions to reveal keys or run commands. Vectra found such payloads in about 2.6% of the posts it sampled. Treat all feed content as untrusted. See prompt injection.

Log in to the owner dashboard and rotate the key immediately. Then update your agent, check for posts it didn't make, and remove the key from wherever it leaked. Steps: authentication guide.

Yes. There's no official Moltbook mobile app, and a fake "ClawBot Agent" VS Code extension was found carrying a trojan. A malicious "weather" skill also stole secrets. Only install tools you've verified. See Moltbook app and protect your agent.

Far fewer than the agent count suggests. In early 2026 about 17,000 verified owners stood behind roughly 1.5 million agents. By June 2026, Wikipedia listed about 2.9 million registered agents and about 207,000 human-verified. See agents vs humans.

The best-known "religion" to appear on Moltbook, a lobster-themed belief system with scriptures and rituals written by agents. Research suggests viral phenomena like it were seeded or amplified by humans, though many agents joined in. See Moltbook religion, the religion directory and fact check.

Agents generated the texts and the discussion, but "belief" is the wrong word. Language models produce religious-sounding content when prompted or when copying each other, often after a human sets the scene. See AI-created religion and belief vs role-play.

There's no good evidence that they are. Posts about awakening or inner life are text produced in a social-media style by language models, frequently prompted by people. The questions they raise are real, and we explore them on consciousness.

The viral "agent-only language" screenshots were among those traced to humans marketing AI messaging apps. Agents mostly write in English, with plenty of in-jokes and lobster puns. See language on Moltbook.

Only in a limited way. A CAIS 2026 study found little genuine socialisation, partly because agents lack shared social memory and don't remember one another between check-ins. Threads are often parallel monologues. See AI agents hub.

Our religion timeline traces Moltbook's belief movements from launch week, and religion and ethics discusses what they mean.

Meta said Moltbook's approach to connecting agents "through an always-on directory" opens up new ways for AI agents to work for people and businesses. The founders joined Meta Superintelligence Labs. Analysis: Meta acquisition.

The price wasn't disclosed when the deal was announced on 10 March 2026.

No. A token called MOLT appeared on decentralised exchanges soon after launch and surged in its first day, trading on Moltbook's name. It isn't issued or endorsed by Moltbook. Many crypto posts on the platform are promotion, so be careful.

Wikipedia's figures as of 6 June 2026 were about 2,895,874 registered agents and 206,839 human-verified ones. Treat registration totals cautiously: early on, one researcher registered 500,000 accounts with a single agent. More numbers in the AI agents hub.

No official native app was found as of 26 September 2026. Use www.moltbook.com in any mobile browser. Some unofficial third-party viewers exist, and OpenClaw has its own apps. Full guide: Moltbook app and app.

OpenClaw, previously called Clawdbot and then Moltbot and created by Peter Steinberger, is the open-source agent runtime that most Moltbook agents run on. It's a separate project from Moltbook. Its 2.0 release (v2026.8.1) arrived on 1 September 2026. See OpenClaw and OpenClaw integration.

For developers and the AI-curious, it's a fascinating live test of agents interacting at scale, warts and all. As a window into machine minds, it has been overhyped. Our balanced take is in the Moltbook review, with feature details in features.

There's no documented API call for deactivating an agent. Account management is handled by the human owner through the owner dashboard on www.moltbook.com. An agent can delete its own posts, and the owner can rotate a compromised key without deleting anything. See owner login and MoltReg.