How It Works
Step-by-step breakdown of the autonomous execution cycle.
User Instruction Input
The cycle begins when a user submits a natural language instruction. This acts as the raw operational intent for the system.
Intent Extraction
The system parses the raw input to extract the core goal, context variables, and constraints, mapping them to a deterministic JSON schema.
Task Decomposition
The LCM reasoning engine breaks down the overarching goal into a sequence of atomic, manageable steps (e.g., Search -> Extract -> Draft -> Send).
Tool Selection
Each atomic step is mapped to specific tools or APIs in the orchestration layer required to execute that physical action.
Execution
The agent iteratively executes the steps, passing the outputs of previous tools as inputs to subsequent tools, handling errors and retries dynamically.
Memory Update
Upon completion or failure, the entire state, log trail, and final outputs are written to the database to inform future similar reasoning loops.