Skip to main content

Open Source & Self-Hosting

SimSwarm is MIT-licensed and self-hostable. The whole stack ships in one repository — the native simulation engine, the FastAPI backend, the Vue frontend, and the GPU job runner. There is no commercial billing layer: authenticated users on a self-hosted instance submit simulations freely.

One codebase, two postures​

A single configuration flag, DEMO_MODE, switches the platform between its two modes:

  • DEMO_MODE=false (default) — full self-hosted platform. Registration and job submission are enabled. This is what you run on your own infrastructure.
  • DEMO_MODE=true — read-only public demo. Signups and job submission are disabled; only the browsable demo gallery and share/read endpoints stay open.

The public site at simswarm.xyz runs with DEMO_MODE=true: it is a read-only window onto a curated set of real simulations, not an account you can sign up for. To actually run simulations, deploy your own instance.

What you keep when self-hosting​

The open-source platform retains everything that produces a simulation result:

  • the swarm engine (simswarm/) — agents, environments, belief dynamics, extraction, graph build, and report generation;
  • authentication, so a self-hosted instance can support multiple users;
  • the full job lifecycle, from seed upload through GPU provisioning to the four result views.

What it takes to run​

Self-hosting brings up the stack with Docker Compose and requires credentials for the external services the pipeline depends on — a GPU provider for the simulation workers, an LLM provider for report generation, and object storage for simulation artifacts. The exact set of environment variables and the bring-up steps are covered in the self-hosting documentation; the project README's quickstart is the fastest path to a running instance.

SimSwarm is released under the MIT license. The simulation engine is fully native, so there is no copyleft engine dependency to inherit.