Home

zLoop

Production-grade framework for bounded autonomous engineering feedback loops.

Version 1.0.0 — stable, tested, ready for deployment.

Overview

zLoop is a vendor-neutral Python framework for building AI agent systems that improve their own work through bounded feedback loops. Every run is constrained by explicit budgets, stop conditions, tool permissions, idempotency rules, independent verification, persistent memory, and human handoff.

Core Lifecycle

DISCOVER → PLAN → EXECUTE → VERIFY → REVIEW → REPAIR → VERIFY → SHIPPED

Quick Start

pip install -e .
python -m unittest src.test_zloop_engine -v
zloop  # Run demo

Features

• Six-dimensional budget enforcement: iterations, repairs, no-progress, tokens, cost, wall-clock
• File-locked concurrent persistence: Safe for parallel agents
• JSON Schema validation: Validated state and results
• Structured logging: Full audit trail
• GPG-signed commits: Tamper-evident history
• Maker/checker separation: Independent verifier and reviewer roles
• Human handoff: Escalation when confidence or budget is low

Documentation

• Installation
• Quick Start Guide
• Core Concepts
• API Reference
• Agent Roles
• Skills
• Examples
• Production Deployment

License

MIT