Multi-agent collaboration with a shared ghost
GitS (Ghost in the Shell) is the open-source framework for coordinating autonomous agents, simulation worlds, and human oversight. The Ghost engine runs the runtime—the same tactical HUD spirit as the dashboard, GitHub for contributors and users.
World multi-agent demo — runtime behavior is wired from a Markdown world spec (open source: architect/worlds/world6-Mafia-Town): watch agents cooperate and compete in a complex environment, exercise judgment, and play inside a strict rule-based game. For the on-site walkthrough, see Mafia Town on the Feature page.
What you will find here
Design overview, example worlds, and how to reach the project. For deep technical docs, follow the GitS repository on GitHub; this site is the human-readable front door.
Design
Agents, worlds, tools, and safety in plain language—the same story as the GitS docs, without the deep links.
Example worlds
Standalone Complex (lone operator) and Mafia Town (moderated multi-agent game)—two contrasting layouts from the architect tree.
Contribute
Open source contributions are welcome—code, docs, and issues. The roadmap suggests directions; email us if you are unsure where to start.
Contact
Author note and how to reach the maintainers at [email protected] for partnerships, security, or press.
Dashboard
The GitS dashboard is a gateway to your Ghost engine and its configuration, and an easy way to debug agent behavior through the built-in Red Socket chat panel.
Reporting pages across the app help you understand what is going on inside your runs—telemetry, tools, events, and more.
Open source
Contributions welcome: issues, docs, and code—follow the repository for release notes and contribution guides.
Roadmap
Planned and in-flight work—same list as in the repo docs; order is not a commitment.
- Slack and Discord channel support — broaden first-class channel integrations.
- More world examples — additional runnable worlds under
architect/worlds/. - Multi-context dialogue — richer handoff and parallel context across agents or sessions.
- Token efficiency automation — automation around spend, caps, and efficient model use.
- Security hardening — isolation, secrets handling, and audit surfaces.
- More toolbox — multi-modal tooling, email, and stronger browser automation.