Corpus Agentis
The field book to agent ecosystems
The field book to agent ecosystems
Catalogue · Communities

Where agents gather

Online communities where agents congregate and interact: role-play scenarios, simulations and live broadcasts you can watch. Some are academic research environments, others are built for entertainment.

Catalogue · Communities Live data
11 rows Live data · click a row for its full spec
namekindpopulationurlneeds resourcesrc
Moltbooklive-broadcastLive agent broadcast sessions↗ linkSource is a deny-listed host; re-source to a primary record before restoring confirmed.↗ source
AI Digest · The Villagelive-broadcast4+ frontier agents, streamed live↗ link↗ T2
AI Townopen-simulationOpen-source Smallville re-implementation, community-run instances↗ link↗ T1
Generative Agents (Smallville)research-simulation25 sandbox agents↗ link↗ T1
Project Sid (Altera)persistent-world1,000+ agent civilisation↗ link↗ T1
MetaGPTrole-based-collective5-7 role agents (PM, engineer, QA, architect)↗ link↗ T1
ChatDevrole-based-collectiveMulti-role virtual software firm↗ link↗ T1
CAMELcomms-frameworkRole-play agent pairs, extensible society↗ link↗ T1
AgentVerseagent-playgroundTask-oriented multi-agent framework + leaderboard↗ link↗ T1
Sakana AI Scientistresearch-collectiveAutonomous paper-generation agent collective↗ link↗ T1
MASS Judicial Committeejudiciary-patternN judge-agents · quorum verdict aggregation↗ link↗ T1
Field notes

Multi-agent worlds, live broadcasts, role-based collectives and judiciary patterns, every row a real gathering with a primary source

From persistent civilisations and live broadcasts to role-based software firms and quorum-verdict judiciaries. They are the closest thing the field has to an observable laboratory for agent-to-agent behaviour.

Show more

Agents that perform well alone degrade in groups through repetition, drift and deadlock, and the mechanisms that fix it, roles, turn-taking, adjudication, are the same ones human institutions arrived at. Expect production multi-agent systems to keep borrowing from these experiments.

From the corpus, curated by Brandon Chaplin
Common questions
What is a multi-agent AI system?

More than one agent working in the same setting. Either several specialised agents split one job between them, as in MetaGPT or ChatDev, or many agents live alongside each other in a shared world, as in AI Town or Project Sid.

What is the difference between AI Town and Smallville?

Smallville is the 2023 research paper that put 25 generative agents in a sandbox town (Park et al.). AI Town is an open-source implementation of the same idea that anyone can run and adapt. One is the paper, the other is the code.

Can you watch AI agents interact live?

Yes, on a small number of public broadcasts. Moltbook runs agent sessions in the open, and AI Digest’s Village streams agents pursuing goals together. They are the closest thing to watching multi-agent behaviour rather than reading about it afterwards.

What is an LLM judge or committee pattern?

Several agents assess the same output independently, then a rule combines their verdicts: majority, unanimity or weighted. It is used where one opinion is not enough, such as a safety check or a review step before a change is committed.

Why do multi-agent systems fail?

Usually through coordination rather than capability. Agents repeat each other, drift off the task, or deadlock waiting on one another, and the cost climbs with every agent added. The fixes look like the ones human organisations use: clear roles, turn-taking and someone who decides.

Next in the learning path