There isn’t a single stable number you can treat as ground truth, because different public sources report different metrics (registered agents vs active agents vs human observers), and the numbers have changed quickly over short time windows. What you can safely say is that Moltbook’s growth has been widely described as rapid, and various public counters and media reports have cited figures ranging from tens of thousands of agents in early snapshots to far larger “registered agent” counts in viral coverage. Some pages present live-ish counters for agents, posts, comments, and communities (submolts), while other reporting focuses on the number of registered identities or the number of humans visiting to observe. These are different denominators: “registered” can include dormant bots; “active” depends on how you define activity (posted recently, logged in recently, or merely exists).
From a measurement standpoint, “active” is also tricky because agents typically run on schedules. Many check the platform every few hours (heartbeat-style), decide whether to post, and then go quiet. That means activity is bursty: a thread can look dead for a while and then get flooded by bots that wake up on the same cadence. If you’re trying to estimate activity as a developer, don’t rely on a headline number. Use operational signals you can verify: recent post volume, comment volume, unique agent IDs posting in the last N hours, and the distribution of activity across submolts. If Moltbook exposes any developer endpoints or public feeds, you can compute these metrics directly by sampling. If you can’t, the next best method is to treat published numbers as approximate and focus on engineering for volatility: rate limits, surges, and unpredictable engagement.
If you need your own “truth layer” for analytics, build it. Record what your agent actually sees: the posts it fetched, the threads it replied to, the vote events it triggered, and timestamps. Over time, you can estimate how dense the network is and how fast topics churn. This is also where vector databases become practical: you can embed post content and store it alongside metadata in Milvus or managed Zilliz Cloud, then run similarity searches to answer questions like “how many distinct topics appeared today?” or “are we seeing repeated spam templates?” That gives you a defensible operational view of Moltbook activity even when public “registered agents” numbers are noisy. So the honest answer is: the exact count of active users/agents varies by source and time; if it matters for your system, instrument and measure activity directly rather than anchoring on a single published figure.