{"repo":"rodmena-limited/stabilize","free":true,"listed":false,"github":"https://github.com/rodmena-limited/stabilize","clone":"git clone https://github.com/rodmena-limited/stabilize.git","description":"Queue-Based State Machine - A lightweight workflow execution engine with DAG-based stage orchestration. Unlike simple task queues (like Celery) or advanced orchestrators (like Highway), Stabilize strikes a balance specifically optimized for high-throughput, stateful DAG execution, making it highly suitable for coordinating autonomous AI agents","language":"Python","stars":86,"topics":["agentic-ai","agentic-workflow","ai-agents","highway","hwe","ocra","orchestration","orchestrator","spinnaker","workflow"],"license":"Apache-2.0","category":"workflow-automation","readme_excerpt":"Stabilize Stabilize is a durable workflow engine for Python. You describe work as a directed graph of stages, and the engine runs it — in parallel where the graph allows, resuming exactly where it left off after a crash, and recording every state transition for audit and replay. It runs embedded in your process on SQLite, or against PostgreSQL for multi-node deployments; there is no separate server or scheduler to operate. The engine was built for reliable orchestration in general, and it turns out to be a particularly good foundation for agentic systems : LLM agents are long-running, make expensive external calls, loop, wait on humans, and must not lose progress or repeat side effects when something fails. Those are exactly the guarantees Stabilize provides. This document is organised around building agentic workflows, from a single model call to a multi-agent system, but the same primitives apply to any pipeline. Build with your coding agent Stabilize ships its own complete reference for AI coding agents. If you work with Claude Code, Cursor, Copilot, or any LLM coding assistant, you don't need to teach it the API — point it at the built-in reference and describe what you want: Then ask your agent, for example: Read stabilize.md , then build a workflow where three agents research a question in parallel, their findings are gathered, a reviewer scores the result and loops back if confidence is low, and a human approves before it is saved. Stream progress as it runs. The refer","default_branch":null,"files":null,"tree":[],"storefront":"/r/rodmena-limited","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/rodmena-limited/stabilize/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}