Spice: open-source decision layer controls AI agent actions before execution

Spice is a new open-source runtime built to solve a critical gap in current AI agent ecosystems: the inability to make good decisions about what to do and when. While tools like Claude Code, Codex, Hermes, and OpenClaw excel at execution, the decision layer has largely remained a manual prompt typed by the user. Spice fills this void by acting as a lightweight 'brain' that sits above these agents. It continuously observes the user's context—tasks, priorities, constraints—and builds a state model. Before any action is taken, Spice simulates possible futures, detects conflicts, and selects the optimal path. It then dispatches the task to the appropriate agent, monitors execution, and reflects on the outcome to improve future decisions. The core loop is perception → state model → simulation → decision → execution → reflection, giving AI systems a structured, auditable decision-making framework.

Spice does not replace agents; it enhances them by adding a layer of reasoning and oversight. The project is fully open-source on GitHub, encouraging community contributions forking and feedback. For professionals building multi-agent systems or complex automation pipelines, Spice offers a way to make agent behavior more predictable and accountable. Instead of each agent acting in isolation based on a single prompt, Spice provides a centralized decision hub that understands context, prioritizes tasks, and learns from past outcomes. This could significantly reduce errors from conflicting commands and improve the efficiency of AI workflows. The open-source nature also allows teams to customize the decision logic to their specific domain, making it a flexible tool for enterprise AI deployments.