- Added Agent Lightning research section to researcher.html with Demo 2 results - Created comprehensive /integrations/agent-lightning.html page - Added Agent Lightning link in homepage hero section - Updated Discord invite links (Tractatus + semantipy) across all pages - Added feedback.js script to all key pages for live demonstration Phase 2 of Master Plan complete: Discord setup → Website completion 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> |
||
|---|---|---|
| .. | ||
| demo1-basic-optimization | ||
| demo2-governed-agent | ||
| demo3-full-stack | ||
| README.md | ||
Agent Lightning + Tractatus Integration Demos
This directory contains demonstrations of how Microsoft's Agent Lightning framework can be integrated with the Tractatus governance framework to create high-performing, values-aligned AI systems.
Core Thesis
Agent Lightning and Tractatus address fundamentally different problems:
- Agent Lightning: "How to do this task better?" (Performance optimization through RL)
- Tractatus: "Should this decision be made at all?" (Governance and values alignment)
Together, they create AI systems that are both:
- High-performing (optimized by AL)
- Values-aligned (governed by Tractatus)
Demo Progression
Demo 1: Basic Optimization (AL Standalone)
Purpose: Show Agent Lightning optimization without governance
Key Learning: AL excels at performance optimization
Files: demo1-basic-optimization/
Demo 2: Governed Agent (AL + Tractatus) ⭐
Purpose: Show Tractatus governing AL-optimized agents
Key Learning: Governance layer prevents values-misaligned decisions
Files: demo2-governed-agent/
This is the killer demo - shows complementarity
Demo 3: Full-Stack Production (Enterprise-Ready)
Purpose: Production-ready AI system with full observability
Key Learning: Real-world deployment architecture
Files: demo3-full-stack/
Architecture
┌──────────────────────────────────────────────────────────┐
│ QUESTION: Should this decision be made at all? │
│ FRAMEWORK: Tractatus (Governance Layer) │
│ FOCUS: Values alignment, human agency, pluralism │
│ ✓ BoundaryEnforcer - blocks values decisions │
│ ✓ PluralisticDeliberator - multi-stakeholder input │
│ ✓ CrossReferenceValidator - instruction adherence │
└──────────────────────────────────────────────────────────┘
↓
[Approved Task]
↓
┌──────────────────────────────────────────────────────────┐
│ QUESTION: How to do this task better? │
│ FRAMEWORK: Agent Lightning (Performance Layer) │
│ FOCUS: Task success, efficiency, optimization │
│ ✓ RL-based optimization │
│ ✓ Training-agent disaggregation │
│ ✓ Framework-agnostic integration │
└──────────────────────────────────────────────────────────┘
↓
[Optimized Execution]
Prerequisites
Agent Lightning
cd ~/projects/agent-lightning
source venv/bin/activate
pip install -e .
Tractatus
cd ~/projects/tractatus
npm install
Python Dependencies (Each Demo)
cd demo1-basic-optimization/
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
Running Demos
Demo 1: Basic Optimization
cd demo1-basic-optimization/
source venv/bin/activate
python task_optimizer.py
Demo 2: Governed Agent ⭐
cd demo2-governed-agent/
source venv/bin/activate
python governed_agent.py
Demo 3: Full-Stack
cd demo3-full-stack/
source venv/bin/activate
python main.py
Use Cases
These demos support three strategic channels:
Channel A: Discord Community
- Share governed agent demo in AL Discord
- Show how Tractatus complements AL
- Propose governance as a contributed feature
Channel B: Academic Publication
- Paper: "Complementary Layers in AI Systems: Governance + Performance"
- Use demos as empirical evidence
- Target: NeurIPS workshop, IEEE, AAAI
Channel C: Public Demonstrations
- Interactive demos at agenticgovernance.digital
- Show live governance intervention
- Educational content for AI safety community
Documentation
- Integration Guide:
~/projects/tractatus/docs/integrations/agent-lightning.md - API Reference:
~/projects/tractatus/docs/api/ - Research Papers:
~/projects/tractatus/research/papers/
Contact
- Project: Tractatus Framework
- Author: John Stroh
- Website: https://agenticgovernance.digital
- Purpose: Preserve human agency over values decisions with plural values context
License
- Agent Lightning: MIT License (Microsoft)
- Tractatus: Apache License 2.0
- Demos: Apache License 2.0
Last Updated: November 2, 2025 Agent Lightning Version: 0.2.2 Tractatus Version: v3.0.2