Spice drops open-source decision layer for AI agents

Spice addresses the critical "orchestration gap" where execution-heavy agents often lack the strategic context to prioritize tasks effectively.

• –Introduces "Decision Cards" to provide a transparent, human-readable audit trail for agentic choices and rejected alternatives.
• –The SDEP protocol enables a clean separation of concerns, allowing developers to swap execution models without rebuilding the core decision logic.
• –Simulation-based reasoning helps mitigate impulsive agent actions by modeling potential outcomes before execution.
• –Lightweight Python implementation facilitates rapid integration into existing agentic frameworks like Claude Code or Codex.
• –It moves the industry closer to "System 2" reasoning for agents, focusing on the "what" and "why" rather than just the "how."