What is SimSwarm
SimSwarm is a swarm-intelligence, agent-based simulation platform for prediction and stress-testing. You upload a seed document describing a situation, set a prediction goal, and run a simulation in which many LLM-driven agents debate, post, follow, and trade across a simulated ecosystem. Instead of asking a single model "what happens next?", SimSwarm lets a population of agents interact over a series of rounds and surfaces the dynamics that emerge from those interactions.
The user flow
- Seed document. Provide the source text the scenario is built from. The seed can optionally be enriched with live web and X/Twitter research before the run.
- Prediction goal. State, in plain language, the question you want answered (for example, forecasting market response or community sentiment over a horizon).
- Run. Agents are derived from the seed and step through the simulation rounds.
- Results. When the run completes you get four views of the output:
- a deep-analysis report with grounded findings,
- an interactive entity graph of participants and their relationships,
- prediction-market data (price charts, trades), and
- a full chat replay of every agent action, round by round.
Who it's for
SimSwarm is aimed at domain experts and researchers who want to explore how a scenario might unfold without writing simulation code. The wizard-driven flow takes a document and a goal; the platform handles agent generation, the simulation loop, extraction, and report assembly.
Try it
A read-only public demo runs at simswarm.xyz, where you can browse a gallery of real simulations and their outputs. SimSwarm is open source and self-hostable; see Open Source & Self-Hosting to run your own instance.