Uses the Pro model tier (NIM_MODEL_CODER / NIM_MODEL_PRO) due to its strong Python coding capabilities, syntax correctness, and ability to handle large token-window accumulative coding logic.
Leverages Langchain's .with_structured_output() ensuring generation of complete, raw, self-contained Python scripts unpolluted by markdown or conversational text.
Follows the PlannerAgent within the primary execution cycle. Acts somewhat like an autonomous Jupyter notebook—extending existing scripts, appending new data-handling/analysis chunks depending on the currently focused plan step, and reacting gracefully to previously failed executions by analyzing stack trace context.
Deployed in the backend engine block. Its output string is immediately digested by the CodeExecutor sandbox.