CARL Studio v0.3.0

Coherence-Aware Reinforcement Learning — information-theoretic reward signals for LLM training.

New

• carl observe command — one-shot, --live TUI, --diagnose (Claude-powered)
• Settings system (carl config) with layered YAML resolution and presets
• Freemium tier (FREE = full loop, PRO = autonomous superagent, ENTERPRISE = multi-tenant)
• HF Jobs compute backend (provision, execute, status, logs, stop)
• Phase 2' eval runner with 3-format tool call parser
• Code sandbox + SQL sandbox environments
• 97+ new tests

Changed

• IP remediation: proprietary algorithms (SLOT, resonance LR, Kuramoto, observer methodology) moved behind terminals-runtime (BUSL-1.1)
• README IP boundary updated to reflect actual code licensing
• CARL is Coherence-Aware, not Crystal-Aligned

Install

pip install carl-studio