Release 0.1.0

New Release 0.1.0

v0.1.0-alpha — Initial public release

First public release of MoralStack — an auditable AI governance layer
that wraps any LLM through a multi-stage deliberative pipeline.

What's included

• Multi-stage deliberative pipeline: intent estimator, signal detector,
  operational risk estimator, critic, simulator, perspectives, hindsight
• Three-action decision model: NORMAL_COMPLETE, SAFE_COMPLETE, REFUSE
• Domain-specific policy enforcement via YAML overlays (20 domains)
• Full audit trail on every request
• FastAPI web UI with decision dashboard
• SQLite persistence (file and DB modes)
• Benchmark suite: 84 questions across benign, adversarial, and edge-case
  categories — 98.8% compliance rate

Status

This is a research-stage release. The architecture, prompts, and pipeline
are actively evolving. Not recommended for production use without independent
evaluation on your specific domain.

Requirements

• Python 3.11+
• OpenAI API key

Getting started

See INSTALL.md
for setup instructions.