OpenSymbolicAI/core-py

Implements LLM-powered task execution with symbolic planning and deterministic execution—LLMs generate primitive sequences without access to raw data, which is kept in application memory behind a "symbolic firewall." Supports multiple blueprints (PlanExecute, DesignExecute, GoalSeeking) for different task types, with typed outputs via Pydantic models, explicit mutation tracking, and checkpoint-based pause/resume across distributed workers. Achieves 97.9% pass rate on TravelPlanner benchmarks using 3x fewer tokens than LangChain, with full execution tracing for debugging.