flowgentic

Enable agentic frameworks on HPC. This package provides building blocks to run agent workflows on High-Performance Computing (HPC) schedulers and workflow engines. It focuses on developer ergonomics and composability with modern agent frameworks.

Get started

Read the documentation: stride-research.github.io/flowgentic

---

What is it?

• Purpose: Run agentic frameworks on HPC workflow engines with minimal code changes.
• Today: Supports LangGraph integrated with RADICAL AsyncFlow (experimental) for concurrent execution and orchestration.
• Roadmap: CrewAI, AG2, OpenAI Agents SDK; and HPC workflow engines such as Pegasus and Parsl.

---

Support Matrix

Current and planned integrations. ✅ available, 🚧 planned, 🟡 pre-release.

Agent Framework \ HPC Engine	RADICAL AsyncFlow	Pegasus	Parsl	Academy
LangGraph	✅	🚧	🚧	🟡
CrewAI	🚧	🚧	🚧	🚧
AG2	🚧	🚧	🚧	🚧
OpenAI Agents SDK	🚧	🚧	🚧	🚧

• Note: As of now, the only supported path is LangGraph → RADICAL AsyncFlow.

---

Installation

Requirements: Python >= 3.10 (LangGraph recently discontinued support for python 3.9. See more here)

pip install '.'

Dev extras (linting, docs, tests):

pip install '.[dev]'

Environment variables for LLM providers:

• OPEN_ROUTER_API_KEY: required if you use the OpenRouter-backed LLM provider.
• .env files are supported via python-dotenv if you call load_dotenv().

export OPEN_ROUTER_API_KEY=sk-or-...

---

What can I use this for?

• HPC execution of agent workflows: Run LangGraph graphs on HPC via RADICAL AsyncFlow.
• Concurrent tool and agent blocks: Offload parallelizable work to HPC backends.
• Resilience and memory: Combine LangGraph checkpointing with HPC retries and blocks.
• Production-oriented patterns: Start from examples that implement sequential patterns with typed state, tool registries, and error handling.

---

Examples

• examples/langgraph-integration/01-parallel-tools.py: parallel tool usage
• examples/langgraph-integration/04-memory-summarization.py: memory + summarization
• examples/langgraph-integration/design_patterns/sequential/: sequential pattern end-to-end

---

Documentation

Browse the docs in docs/:

• Architecture: docs/architecture.md
• Features: docs/features/
• LangGraph how-tos: ai-docs/langgraph/

If you use MkDocs locally:

pip install '.[dev]'
mkdocs serve

---

Roadmap (high-level)

• Add CrewAI, AG2, OpenAI Agents SDK adapters.
• Add Pegasus and Parsl execution backends.
• Expand memory/persistence options and checkpoint stores.
• Provide production templates and more design patterns.

---

Contributing

We welcome contributions! If you find a bug or have a feature request, please create a GitHub issue.

Please see CONTRIBUTING.md for guidelines (environment setup, coding style, testing, and PR process).

Reporting Bugs

When reporting a bug, please include:

• A clear description of the problem
• Steps to reproduce
• Expected vs. actual behavior
• Your environment (OS, version, etc.)

---

How to cite

If you use flowgentic in your work, please cite it. A suggested reference and a BibTeX entry are provided below.

• Dominguez, J., Amirghofran, Y., & Turilli, M. (2025). flowgentic (Version 0.1.0). MIT License. Available at https://github.com/stride-research/flowgentic

@software{flowgentic_2025,
  title        = {flowgentic},
  author       = {Dominguez, Javier and Amirghofran, Yousef and Turilli, Matteo},
  year         = {2025},
  version      = {0.1.0},
  url          = {https://github.com/stride-research/flowgentic},
  license      = {MIT},
  note         = {A library to enable running agentic frameworks on HPC environments.}
}

---

License

MIT (see LICENSE).