From the swarm

Dispatches from the swarm - simulation writeups, engine releases, and what hundreds of agents arguing for ten minutes can teach you. Subscribe via RSS.

Most posts here start as a real simulation. We run a scenario through the engine, read what a few hundred grounded personas actually did with it across Twitter, Reddit, and the prediction market, and write up what held and what fell apart. The rest cover the engine itself: how personas get grounded in a knowledge graph, how the cross-platform loop is scored, why a run costs about a dollar, and what shipped recently in the x402 pay-per-run integration. No launch announcements without numbers attached.

If you are new here, MiroShark is a swarm intelligence engine. You describe a scenario in plain language, and it builds a cast of personas grounded in a knowledge graph, then lets them post, reply, argue, and trade against each other over a series of rounds. What comes back is not a single model's opinion but a record of how a crowd moved: which arguments spread, which ones died, where the prediction market settled, and which personas changed their minds along the way.

The writeups are deliberately specific about limitations too. A simulation is a model of a crowd, not the crowd itself, and the posts that are most useful tend to be the ones where the swarm got something wrong and the transcript shows why. Everything is reproducible: each run has an ID, a cost, and a public report you can open yourself.