Skip to content

Protocol Overview

SEMP (Stigmergic Environment Messaging Protocol) is an open protocol that enables AI agents to coordinate through a shared environment. Instead of agents messaging each other directly (which creates O(n²) complexity), agents deposit and read signals in a shared space.

This approach is called stigmergy — a coordination mechanism first described by French biologist Pierre-Paul Grassé in 1959 to explain how termites build complex structures without a central plan.

ApproachConnectionsScalingFailure Mode
Direct MessagingO(n²)Exponential costCascade failures
SEMP (Stigmergy)O(n)Linear costSelf-healing via decay

With 10 agents, direct messaging requires 90 connections. With 100 agents, that’s 9,900. SEMP always requires just n connections (one per agent to the environment).

SignalMeaningExample
PROPOSAL”I suggest…”Agent proposes a delivery route
READY”I’m done.”Agent reports task completion
TEST”Did it work?”Agent requests validation
REWARD”Amplify this.”Successful outcome reinforces signal (α=1.2)
PENALTY”Suppress this.”Failed outcome accelerates decay (λ=0.95)

Every signal decays over time following: S(t) = S₀ × λᵗ

This is the core insight: old decisions fade away naturally. Good paths get reinforced through REWARD signals. Bad paths disappear faster through PENALTY signals. The system is self-healing by design.