Quickstart

Head to the subscription page and create an account. Agent Orchestrator is usage-based — you only pay for the tokens and executions you use.

Once logged in, open Agent Orchestrator from the sidebar.

Navigate to Workflows and click New Workflow. Use the built-in CodeMirror editor — or pick a template from the gallery — to define your workflow:

# Market Research Report

## Inputs
- topic: The subject to research (string, required)

## Agents

### Researcher
Expert at gathering and synthesizing information from multiple sources.
Model: Sonnet
Tools: rag-engine

### Writer
Skilled at writing clear, structured reports.
Model: Sonnet

## Steps

### Research
Search the knowledge base for information about {inputs.topic}.
Find key facts, trends, and insights.

### Write Report
Using the research: {steps.research.output}

Write a comprehensive report on {inputs.topic}.
Include an executive summary, key findings, and recommendations.

## Output
steps.write_report.output

See the Workflow DSL Reference for the full syntax.

Click Save. The orchestrator compiles your markdown into a structured execution plan using Claude. You'll see a preview of the compiled plan showing agents, steps, and dependencies.

If there are issues (missing agents, circular dependencies), the compiler will flag them with notes so you can fix them before running.

Click Run on the workflow page. Fill in any required inputs (like topic) and confirm.

The execution page shows real-time progress:

• A segmented progress bar showing which step is running
• Step-by-step results — expand each step to see its output
• Token usage per step and total
• A Stop button if you need to cancel mid-run

When the workflow completes, the final output renders at the bottom of the execution page. Content is auto-detected — markdown renders as formatted prose, JSON is syntax-highlighted, and code blocks display with language hints.

Click Download to save the output as a file (.md, .json, etc. based on content type).