langflow-ai/langflow
Langflow is a powerful open-source platform for building, testing, and deploying AI-powered agents and workflows with visual authoring, multi-agent orchestration, and API integration capabilities.
Awesome AI Agents › AI Workflow Orchestrators
maze-agent/Maze
Maze is a distributed runtime for LLM agent workloads that treats an agent program as a workflow graph rather than a single process. Workflows are defined with a Python decorator, submitted as portable maze.workflow/v1 specifications, or extended at runtime through persisted dynamic runs that append tasks while execution is in progress. Every client path, including the Python SDK, a LangGraph adapter, the browser-based Workbench and application specifications, converges on the same contract and the same core run identity, so there is one execution and observability surface instead of several. The scheduler maintains separate gpu, cpu and io queues so that one resource class cannot block another, and it supports both first-come-first-served ordering and the HACS algorithm described in the project research paper, which was accepted to SC26. Node placement is a separate concern from ordering: a placement strategy picks a registered worker with suitable resources. Model execution is managed by the cluster itself, discovering local checkpoints, deploying reusable vLLM or Transformers instances on demand, routing model tasks to them, reserving GPUs and scaling in and out; a task waiting for a model instance is held in an explicit wait state rather than occupying a dispatchable queue slot. Runs are durable across process restarts and retain task state, structured errors, events, logs, retries, timeouts, cancellation, placement records and content-addressed artifacts. The Workbench adds a DAG editor, catalogs of reusable tasks and workflows, server-side workspaces with file management, validation, run inspection and cluster operations. Ray provides the distributed execution layer underneath, while Maze contributes workflow contracts, resource semantics, durable state, scheduling, model lifecycle and operational APIs. It installs from PyPI and suits teams running multi-step agent pipelines on shared GPU clusters.
https://github.com/maze-agent/Maze
Also in AI Workflow Orchestrators
stablyai/orca
Orca is an AI Orchestrator that provides an integrated development environment for running and managing parallel AI agents across various platforms, including desktop and mobile.
calesthio/OpenMontage
OpenMontage is an open-source, agentic video production system that transforms AI coding assistants into full video studios for automated video creation from text prompts.
stanfordnlp/dspy
DSPy is a framework that enables programming language models with compositional Python code to build modular AI systems and optimize their performance beyond traditional prompting methods.
kestra-io/kestra
Kestra is an open-source, event-driven orchestration platform for data, AI, and infrastructure workflows, enabling declarative and code-based automation with a rich plugin ecosystem.
humanlayer/12-factor-agents
12 Factor Agents is a comprehensive framework outlining principles for building reliable, scalable, and production-ready LLM-powered AI software agents.
yoheinakajima/babyagi
BabyAGI is an experimental framework for building self-building autonomous agents using a function-centric approach with comprehensive management, logging, and a dashboard interface.
teableio/teable
Teable is an AI spreadsheet platform that integrates data management, AI workflows, and app building on a PostgreSQL backend, enabling users to create custom business applications.