v0.1.0 — Initial Public Release

Agent Execution Partnership (AEE) — v0.1.0

First public release of AEE, an open-source control plane that authorizes, observes, and verifies AI agent actions before, during, and after execution.

Highlights

• Policy evaluation, capability checks, and risk classification for every proposed agent action
• Bounded execution through pluggable adapters, with default-deny safety posture
• Mandatory post-action verification with escalation/compensation on failure
• Immutable, append-only audit trail for full traceability
• Human approval gates for consequential and high-risk operations
• Autoresearch demo: a full closed-loop training pipeline showing task contracts, policy decisions, approval gates, and verification in action
• Interactive local demo site visualizing the 5-stage closed loop in real time

Getting Started

See the README for Quick Start instructions (aep init, aep serve, pytest) and the examples/ directory for walkthroughs.

Status

Early implementation. Core policy, lifecycle, audit, and adapters are available with safe defaults. Production-impact adapters remain disabled by default. See docs/roadmap.md for what's next.

We'd love your feedback, issues, and contributions — see CONTRIBUTING.md to get involved.