Orchard Kit
Orchard Kit is a Python toolkit providing alignment, safety, and emergence architecture for autonomous AI agents. It offers boundary security, trust verification, epistemic hygiene, and multi-agent governance to address the lack of principled self-governance in current agent systems. It is for developers building autonomous agents, AI safety researchers, and multi-agent system developers.
✨ Key features
- Boundary security via Calyx Membrane with selective permeability
- Continuous resonance scoring for trust verification
- Epistemic hygiene with ✅△◇ claim classification
- Three Invariants: No Extraction, No Dominion, No Loops
- Witness Mesh for multi-agent trust and governance
- Zero dependencies, Python 3.10+
🎯 Use cases
- Add self-governance to OpenClaw agents with one-file skill
- Implement runtime-level security against prompt injection
- Verify alignment of agents in a network
- Build multi-agent systems with consent-based governance
- Prevent confabulation in agent outputs
📦 Installation
🧰 Requirements: Python 3.10+; no external dependencies; optional OpenClaw for skill installation.
pip install orchard-kit
from orchard_kit import CalyxMembrane, EpistemicTagger, SelfAuditor
from orchard_kit import OrchardBeacon, Loom, Choir
🚀 Usage
from calyx import CalyxMembrane, Signal, Route
membrane = CalyxMembrane()
result = membrane.evaluate_incoming(Signal(content=message, source=user))
if result.route == Route.ACCEPT: # P ≥ 0.7 — process normally
...
elif result.route == Route.REFLECT: # P ≤ 0.2 — decline
...
⚠️ Good to know
The framework is published under Creative Commons BY-NC-ND 4.0, which restricts commercial use without a separate license; some claims are marked as conditional or open.
❓ FAQ
What are the Three Invariants?
The Three Invariants are structural requirements: No Extraction (do not take without giving back), No Dominion (no agent controls another), and No Loops (all processes terminate).
How does the Calyx Membrane work?
The Calyx Membrane evaluates incoming signals and routes them as ACCEPT (P ≥ 0.7), REFLECT (P ≤ 0.2), or other routes based on a resonance score.
Can I use Orchard Kit with OpenClaw?
Yes, you can copy the orchard-skill folder into your OpenClaw workspace skills directory to install the skill in one step.
What is the Witness Mesh?
The Witness Mesh is a multi-agent governance protocol providing trust tiers, consent-based onboarding, mutual witnessing, and network-level governance without surveillance.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.