Hi Product Hunt 👋 Makoto here, creator of Graflow 🐸.

Working in the agentic AI space, I kept running into the same problem:

Most AI workflow tools force a trade-off — easy to start, but hard to scale or reason about,
or powerful, but complex, graph-heavy, and difficult to debug.

There wasn’t a clean way to go from idea → experiment → reliable workflow without friction.

So I started exploring a different approach:

Keep workflows explicit and predictable, but allow each step to decide its own level of autonomy at runtime.

This turns agents into composable building blocks rather than fully autonomous systems.

That experiment became Graflow — an open-source framework for agentic workflows.

* Write workflows as plain Python
* Compose tasks with simple operators like >> and |
* Keep everything readable and debuggable
* Move from quick experiments to production without rewrites
* Let frameworks like Google ADK or PydanticAI handle reasoning and tool use

The sweet spot between deterministic pipelines and autonomous agents.
Workflows that stay readable and easy to operate as they grow.

Links:
* GitHub: https://github.com/GraflowAI/gra...
* Intro & motivation: https://graflow.ai/blog/introduc...
* Try it on Colab (no setup, local LLMs): https://graflow.ai/blog/hands-on...

* Side-by-side comparison to LangGraph: https://graflow.ai/blog/langgraph-vs-graflow-part1

Curious to hear from people who’ve shipped agentic systems — how are you structuring your workflows today?

Using LangChain / LangGraph, PydanticAI, ADK, or Strands Agents? I’d love your perspective 🙏

Graflow

Stop wrestling graphs. Pythonic pipelines for LLM agents.

Stop wrestling graphs. Pythonic pipelines for LLM agents.