axiomesh 4.2.0

Environment Design Pattern (EDP) + Model Environment Protocol (MEP) — Causal, contextual, multi-agent AI architecture

axiomesh

Environment Design Pattern (EDP) + Model Environment Protocol (MEP)

OneOrigine / ImperialSchool Research — I.S. License

"A perfect balance between rationality and probability is necessary to fully leverage the logic of artificial intelligence." — Seikatsu-One

Install

pip install axiomesh
pip install axiomesh[ollama]     # + local Ollama
pip install axiomesh[all]        # + all LLM providers

Quick start — Multi-agent CLI

# Interactive school scenario (admin + teacher + accountant)
axiomesh-cli --demo
axiomesh-cli --scenario school --model gemma3:12b
axiomesh-cli --scenario bank
axiomesh-cli --scenario drone

# Run a task file
axiomesh-cli --task school_demo.axm

# Generate a demo task file
axiomesh-cli --gen-task --scenario school

Chat mode commands

Command	Description
/agents	List all active agents
/focus admin	Direct chat to the admin agent
/focus all	Broadcast mode (all agents)
/broadcast <msg>	Send goal to ALL agents simultaneously
/ctx	Show context + harmony-ranked actions
/switch fin accountant	Switch accountant to financial context
/task file.axm	Run a task file
/spawn analyst fin	Spawn a new agent
/impact	Session impact matrix
/causal	Causal graph (DOT)
/why student.enroll	Causal trace
/savoir	SAVOIR knowledge base
/export	Export session JSON

Task files (.axm)

# school_demo.axm
@agent admin
@context admin
@goal Add student Alice Chen

@parallel
@agent admin @context enrl
@agent admin : Enroll student STU001 courseId=CRS001

@agent accountant @context fin
@agent accountant : Process fee payment studentId=STU001
@end

Python API

from axiomesh import Environment, Context, Action, Reaction, Element
from axiomesh import MepGateway, SenseVector, Circumstance, EnvironmentKind, ContextKind

env = Environment("MyEnv", EnvironmentKind.REACTIVE)
ctx = env.create_context("Main", ContextKind.SEMANTIC,
      basis=SenseVector.normative("operations", 0.9))

# Add action
async def handler(actor, payload, ctx, frame):
    return Reaction.ok("my.action", f"Done: {payload}")

ctx.reg(Action("my.action", "command", "Do something",
               SenseVector.normative("action", 0.9), handler=handler))

# MEP gateway
gw      = MepGateway(env)
session = gw.connect("agent-1")
envelope = gw.build_envelope(session, actor, ctx)
reaction = await gw.dispatch(session, actor, "my.action", {}, ctx)

Mathematical Foundations

• Central equation: E_{t+1} = 𝔘(E_t, 𝔯(x,c,Σ, 𝔄(x,c,Σ,Γ,Ψ(D,c,Σ))))
• Harmony: H = α·cos(A,C) + β·cos(A,S) + γ·cos(R̂,R) − δ·D
• Admissibility: Adm(a) = ∧_{γ∈Γ_a} γ(E,c,Σ,x)
• Optimum: a* = argmax_{a∈Avail} H(a,c,s)