Scion#

Scion is an experimental multi-agent orchestration testbed by google-cloud-platform. It manages concurrent LLM-based agents running in containers across local machines and remote clusters.

What It Is#

A “hypervisor for agents” — infrastructure for running, isolating, and managing LLM agent processes. It is explicitly not a full multi-agent framework. Components like agent memory, chatrooms, and task management are treated as orthogonal concerns.

Architecture#

Scion follows a Manager-Worker pattern:

• CLI (scion): Host-side orchestrator managing agent lifecycle and groves
• Agents: Isolated containers running LLM software via harness adapters

Two operating modes:

• Solo Mode: Local-only, zero-config. CLI manages agents directly via Docker/Podman/Apple Container.
• Hosted Mode: Distributed. A centralized hub coordinates state and dispatches to runtime-brokers.

Key Commands#

scion init                          # Initialize project
scion start <name> "<task>"         # Launch an agent
scion attach <name>                 # Interactive session
scion logs <name>                   # View output
scion resume <name>                 # Restart preserving state
scion list                          # Show all agents
scion delete <name>                 # Clean up
scion templates [create|list|show]  # Manage templates

Supported Harnesses#

Configuration#

• Global: ~/.scion/settings.yaml
• Project: .scion/settings.yaml
• Templates: .scion/templates/ or ~/.scion/templates/

Written In#

Go. Key packages: pkg/agent/, pkg/harness/, pkg/runtime/, pkg/hub/, pkg/runtimebroker/.

Philosophy#

• Less is more — hypervisor, not full stack
• Isolation over constraints — --yolo mode + container isolation
• Interaction is imperative — humans and agents collaborate
• Diversity yields quality — mix models, harnesses, configurations
• Action over pondering — testbed for experimentation

See Also#

• agent
• grove
• hub
• harness
• template
• runtime
• agent-state-model
• multi-agent-orchestration