AxiomMesh SDK

EDP + MEP + SAVOIR for context-aware, certainty-aware, federated multi-agent systems.

AxiomMesh is a release-ready SDK for building agents that operate inside structured environments rather than free-form prompt loops. It combines:

• EDP — an environment design runtime with contexts, rules, actions, reactions, mission planning, and mathematical world export
• MEP — the Model Environment Protocol, a JSON-RPC-based protocol surface for envelopes, explanation, planning, execution, federation, provenance, health, and resilience
• SAVOIR — a certainty layer that separates what is observed, verified, derived, stale, contradicted, or shared
• Drone SDK — a reference implementation for drones, swarms, internal sub-agents, and federated mission execution

Positioning

This SDK is designed for systems where the world matters:

• drones and swarms
• embodied AI
• robotics middleware
• multi-agent coordination
• internal subsystem orchestration inside one machine
• federated execution across environments

The core idea is simple:

the agent should not guess the world from scratch every turn. it should receive a structured environment, a constrained action surface, and a certainty-aware state.

What makes this stack different

EDP

EDP turns the environment into an executable decision surface:

• contexts deform the available action space
• rules can be hard or soft
• roles and situation can limit action visibility
• plans can be built locally, cooperatively, or across environments
• the environment can be exported as matrices, graphs, adjacency structures, and factor-like constraints

MEP

MEP is not a prompt wrapper. It is not an MCP overlay either. It is the protocol layer between agents and structured environments. Its scope is broader than tool invocation:

• world state and certainty transport
• constrained action surfaces
• explanation and why-not semantics
• multi-agent and multi-context coordination
• federated execution across environments
• provenance, replay, resilience, and recovery

It exposes:

• context envelopes
• WHY / WHY-NOT explanations
• health and protocol introspection
• planning and mission preview
• multi-agent bindings and shared envelopes
• negotiation, execution, rollback, recovery, and rerouting
• federation across multiple environments
• chained provenance

SAVOIR

SAVOIR is the certainty engine. It distinguishes between:

• observed
• verified
• derived
• stale
• contradicted
• shared

This lets the runtime reason on a validated operational state instead of collapsing every fact into probability.

Repository layout

src/
  edp_core/      # semantic layer, contracts, rules, runtime, missions, multi-agent coordination
  savoir_core/   # certainty, evidence, local/shared knowledge, mesh propagation
  mep_core/      # protocol surface, gateway, federation, JSON-RPC, registry/spec
  drone_sdk/     # drone/swarm reference implementation
  edp/           # compatibility imports
  mep/           # compatibility imports
  savoir/        # compatibility imports

Install

Local development:

pip install -e .

Quickstart

Run the drone demo:

PYTHONPATH=src python -m drone_sdk.demo

Run the full test suite:

PYTHONPATH=src python -m unittest discover -s tests -v

MEP vs MCP

MEP is not implemented as a thin layer over MCP.

The difference in design intent is structural:

• MCP is centered on exposing resources, prompts, and tools to language-model applications
• MEP is centered on exposing environments, contexts, action surfaces, certainty states, causal traces, and federated multi-agent execution

In other words:

• MCP helps an application call capabilities
• MEP helps an agent operate inside a constrained, evolving world

MEP can be bridged to other ecosystems later, including MCP-style surfaces, but its native model is environmental, causal, and multi-agent rather than merely tool-oriented.

Protocol surface

The runtime can expose its protocol shape directly:

• mep.spec
• mep.spec.schema
• mep.spec.markdown
• mep.method.describe
• mep.health

Generated protocol artifacts are included in this repository:

• MEP_SPEC.md
• MEP_PROTOCOL_SCHEMA.json
• MEP_METHOD_CATALOG.json
• MEP_METHOD_CATALOG.md

Current status

• SDK version: 1.0.1
• MEP version: 2.0.0
• Status: release-ready public baseline

Publication assets included

• LICENSE
• NOTICE
• CHANGELOG.md
• RELEASE_NOTES.md
• CONTRIBUTING.md
• SECURITY.md
• CODE_OF_CONDUCT.md
• CITATION.cff
• MEP_SPEC.md
• MEP_PROTOCOL_SCHEMA.json
• MEP_METHOD_CATALOG.json
• tools/export_protocol_catalog.py

License

Apache-2.0. See LICENSE and NOTICE.

Publishing

The repository includes GitHub Actions workflows for CI and tagged publishing.

• .github/workflows/ci.yml runs the test suite on pushes and pull requests
• .github/workflows/publish.yml builds and publishes on tags like v1.0.1

Before a real release, set your repository URLs in pyproject.toml and configure a PyPI Trusted Publisher for the GitHub repository.