Stigmergy: An agents framework built on ECS principles (github.com)

Stigmergy is a new agent framework built on entity-component-system (ECS) principles that models coordination the way termite mounds or ant trails do: through local actions on a shared environment rather than central control. It replaces static class hierarchies with runtime-defined entities (cryptographically strong 32‑byte IDs prefixed "entity:"), components (named types paired with JSON Schemas that support recursion, oneOf unions, and full validation), and systems (agent definitions in Markdown with YAML-like frontmatter). Systems declare capability-based component access (read/write/execute) and express interest via bid expressions—boolean condition and value clauses—evaluated against component data; an auction selects the highest bidder to act. Component schemas can be registered at runtime, validated on create/update, and generated from Rust types via a derive macro, combining runtime flexibility with strong structural guarantees. For AI/ML and multi-agent engineering, stigmergy is significant because it enables emergent, composable behaviors without recompilation or tight coupling: new behaviors arise by attaching components or adding systems that bid on state, facilitating rapid experimentation in simulations, game AI, distributed robotics, or orchestration of model-based tools. The fine-grained capability model and schema validation provide safety and auditability across distributed clients, while the bidding/auction protocol supplies an explicit, declarative coordination primitive for prioritizing conflicting actions—making stigmergy a practical substrate for scalable, data-driven agent ecosystems.