Developer Docs
architecture · spec · sdk
Core Architecture5 min read
DAG Goal Decomposition
Parallel and dependent step execution rules, retries, and fallback logic.
Directed Acyclic Graph (DAG) Execution
#Complex goals often require sequential dependencies (e.g., fetch data before analyzing it) or independent steps that can run in parallel (e.g., scrape two sites at once).
Example DAG Structure
#For the goal: "Fetch portfolio holdings over MCP, summarize Nvidia news via A2A, and take a screenshot of the dashboard."
json
{
"goal_id": "goal-9842",
"nodes": [
{
"id": "step-1",
"protocol": "MCP",
"agent_did": "did:orcha:agent:finance-dashboard",
"dependencies": []
},
{
"id": "step-2",
"protocol": "A2A",
"agent_did": "did:orcha:agent:web-scraper",
"dependencies": []
},
{
"id": "step-3",
"protocol": "COMPUTER_USE",
"agent_did": "did:orcha:agent:desktop-screenshot",
"dependencies": ["step-1", "step-2"]
}
]
}Parallel Dispatch
Steps 1 and 2 execute concurrently in parallel worker threads. Step 3 automatically waits until both Step 1 and Step 2 output verification checks pass.