Core Concepts
Agent Declaration
Define named agents with specific tasks
@writer draft a blog post
@analyst review traffic logs
Agent handles must start with @ followed by letters, numbers, underscores, or hyphens
Variable References
Use template variables in your workflows
=== VARS
topic: AI Safety;
=== END
@writer write about {{ topic }}
Define variables in VARS section and reference with mustache syntax
Basic Execution Patterns
Independent Execution
Default behavior - agents run independently
@agent1 list chocolate types
@agent2 describe chocolate history
No data flow between agents. Execution order not guaranteed.
Sequential Chain
Chain agents with the pipe operator
@researcher gather facts |
@writer create article
Output of left agent becomes input to right agent.
Parallel Execution
Multiple agents share the same task
@designer @developer
create user registration form
Both agents receive the same task and execute in parallel.
Script Structure
Complete Script Example
A well-formed Orchestra ML script with all sections
=== AGENTS
researcher: {persona: "academic researcher"
writer: {persona: "technical writer"
=== END
=== VARS
topic: Machine Learning;
=== END
@researcher gather information about {{ topic }} |
@writer create comprehensive article
Validation Rules
Valid Patterns
@writer create post
@agent_1 do task
@my-agent process data
Invalid Patterns
@writer: draft post
@writer
@1writer task
- • No colons after agent names
- • Agents must have tasks
- • Can't start with numbers