Coordinate multiple specialized agents to accomplish complex goals. Build multi-agent teams that collaborate sequentially, share context, and execute sophisticated workflows.
Orchestras coordinate multiple agents to accomplish complex goals through sequential execution. Unlike individual agents working in isolation, orchestras enable true multi-agent collaboration.
Each agent builds upon the previous agent's output in a coordinated pipeline.
Results from one agent automatically become input for the next agent.
Workspace directories store intermediate results and artifacts.
Complex multi-step processes with task dependencies and priorities.
Load definitions from YAML
Determine execution order
Run assigned agent
Output becomes input
Save to workspace
| Tier | Execution Strategy | Parallel Tasks | Advanced Techniques |
|---|---|---|---|
Oracles & Genies | Sequential only | 1 | Basic orchestration |
Higher Tiers | Parallel, Mixed, Hierarchical | Multiple | Kubernetes orchestration |
Pre-built orchestra patterns for common multi-agent workflows.
Complete SDLC workflow from implementation to testing and deployment
super orchestra run sdlc --goal "Build user auth"Research, write, and polish comprehensive content pieces
super orchestra run content --goal "Create ML guide"Collect, analyze, and generate insights from data
super orchestra run analysis --goal "Analyze survey"metadata:
name: "Software Development Lifecycle"
id: "sdlc"
version: "1.0.0"
orchestra:
workspace:
type: local_fs
path: "./orchestra_workspaces/sdlc"
execution:
strategy: "sequential"
max_parallel_tasks: 1
task_timeout_seconds: 300
enable_metrics: true
enable_trace: true
agents:
- developer
- devops_engineer
- qa_engineer$ super orchestra create <name>Create new orchestra from existing agents$ super orchestra listList all orchestras in project$ super orchestra run <name> --goal "..."Execute orchestra with specific goal$ super orchestra run <name> --verboseRun with detailed output$ super orchestra run <name> --dry-runValidate configuration without runningsuper agent compile --allCompile all agents first
super agent evaluate <agent>Evaluate baseline performance
super agent optimize <agent>Optimize individual agents
Good: Specific & Actionable
"Build a secure user authentication system with JWT tokens, password hashing, and email verification"
❌ Avoid: Too Vague
"Make a login system"
Design logical task dependencies for proper workflow:
Monitor execution and handle failures:
Available in higher tiers beyond Oracles and Genies
Run multiple tasks simultaneously for faster completion.
Combine parallel and sequential execution in task groups.
Scale orchestras with Kubernetes for enterprise workloads.
Coordinate multiple specialized agents to accomplish complex goals. Start with our step-by-step tutorial.