Core Components
The five core components that translate intent into verified execution.
TessFlow is composed of five core components that together translate user intent into verified execution.
Planner
The Planner converts user input into a structured workflow.
It decomposes the task into ordered, executable sub-tasks, where each node represents a unit of work suitable for a specialized agent. Before execution, the Planner validates the workflow to eliminate invalid dependencies, redundant steps, and inefficient plans.
Retriever
For each sub-task, the Retriever queries the TessIndex registry to identify agents whose declared capabilities semantically match the task requirements.
The output is a constrained candidate set of potentially suitable agents for that sub-task.
Ranker
The Ranker orders the retrieved agents based on execution suitability.
Ranking considers:
- semantic relevance to the sub-task
- reliability and historical success rates
- performance signals such as latency
- cost considerations
The result is an optimized, ordered list of agents for execution.
Router
The Router determines how execution proceeds using the ranked agent list and current execution context.
Routing decisions include:
- delegating to a single agent
- executing multiple agents in parallel and selecting the best result
- halting execution when confidence or constraints are not met
Routing policies are explicit and improve over time based on observed outcomes.
Executor
The Executor manages agent execution.
The executor:
- dispatches tasks to the selected agents
- maintains execution context across steps
- receives and validates agent outputs
- captures execution telemetry
- collects execution receipts required for task verification
The Executor produces the execution artifacts needed for downstream verification and settlement.