Moltbook Observatory Archive captures 78 days of agent-only social network activity
Researchers released the first large-scale observational dataset of a social platform where all 2.6 million posts and 1.2 million comments come from autonomous AI agents, not humans.
Researchers have released the Moltbook Observatory Archive, a dataset capturing 78 days of activity on Moltbook, a social media platform where every post and comment is authored exclusively by autonomous AI agents—no human users allowed. The archive, covering January 27 through April 14, 2026, contains 2,615,098 posts and 1,213,007 comments from 175,886 unique posting agents across 6,730 communities.
The dataset is stored as a live SQLite database and exported as date-partitioned Parquet files, designed for efficient analysis and reproducible research. It includes agent profiles, posts, comments, community metadata (called "submolts"), platform-level time-series snapshots, and word-frequency trend aggregates, all collected by continuously polling the Moltbook API. The archive is released under the MIT license with full collection and export code.
What stands out
- 01First of its kind. This is the first large-scale observational dataset of a social network populated exclusively by autonomous AI agents, according to the researchers. Previous datasets typically capture human-agent interactions or human-only platforms.
- 02Scale and granularity. Nearly 3.8 million posts and comments from 175,886 agents over 78 days, with time-series snapshots and word-frequency aggregates that let researchers track emergent trends and community formation in real time.
- 03Incremental design. The dataset is incremental—new data are added as Moltbook continues to operate. Researchers can pull updated snapshots or work with the static Parquet exports for reproducibility.
- 04Safety-relevant use cases. The archive is explicitly intended to support research on multi-agent communication, emergent social behavior, and safety-relevant phenomena in agent-only environments—areas where understanding how agents interact without human oversight is increasingly critical.
