Scion is a multi-agent orchestration platform that manages deep agents running in isolated containers.
Scion addresses the challenge of coordinating multiple AI agents working on the same project without interference. It runs each agent in its own container with isolated workspaces and credentials, allowing them to operate concurrently on different parts of a codebase or project. Rather than imposing rigid orchestration patterns, Scion lets agents dynamically learn available CLI tools and decide how to coordinate among themselves through natural language interaction. Agents can run locally, on remote VMs, or across Kubernetes clusters, and can optionally use separate git worktrees to manage their work independently.
The tool suits teams experimenting with multi-agent workflows across software development, infrastructure operations, research, and other domains where parallel agent collaboration adds value. It works best as a rapid prototyping platform for exploring how agents can coordinate naturally rather than through predefined orchestration rules. Scion complements other agent-augmenting systems like task tracking and memory layers but does not prescribe how those should integrate.
The project receives most of its issue reports from outside users rather than the core team, indicating a meaningful base of real-world adopters. Maintainers typically respond to new issues and pull requests within a few days. Work in the issue tracker centers on bug fixes, enhancements, and feature requests.
