Developer Docs
architecture · spec · sdk
Core Architecture6 min read
SuperAgent Execution Pipeline
How one natural language goal transforms into a verified, protocol-dispatched DAG graph.
SuperAgent Execution Pipeline
#When a user submits a goal to Metaorcha, the SuperAgent acts as the top-level executor. It does not perform domain tasks directly; instead, it manages the workflow state machine.
SuperAgent Execution Pipeline Diagram6-Phase Harness
- 1. PLANGoal → DAG
- 2. ROUTEVector DID Search
- 3. DISPATCHProtocol Handler
- 4. VERIFYSemantic Judge
- 5. NORMALIZEContext State
- 6. RENDERCanvasKit UI
Pipeline Phases
#- Goal Decomposition (Planner): Analyzes the prompt and breaks it into discrete sub-tasks with dependency links (DAG).
- Agent Discovery & Routing: Queries the Registry using vector search and capability filters to match sub-tasks to registered DIDs (
did:orcha:agent:*). - Protocol Dispatch: Routes each step to its corresponding protocol handler (
MCPHandler,A2AHandler,ComputerUseHandler,LangGraphHandler, etc.). - Verification & Judging: Passes agent outputs through the
ExecutionObserverfor schema compliance and semantic quality check. - State Normalization: Aggregates verified outputs into a structured context window.
- CanvasKit Rendering: Converts aggregated data into a declarative JSON layout for frontend visualization.