DEV Community

fresed
fresed

Posted on

fagent: memory-first AI agent runtime for long-running, interrupted workflows

Hi everyone! 👋

I’m a beginner developer fascinated by AI, neural networks, and agents. I built fagent, a memory-first AI agent runtime, because I noticed that most agent runtimes handle short tasks well, but struggle with long-running or interrupted workflows. Agents often forget context, previous decisions, or dependencies once you restart them.

What makes fagent different:

  • Layered memory – multiple memory types (file, vector, graph, task/workflow) working together.
  • Graph-based recall – retrieve not just text, but the relationships between tasks, blockers, and decisions.
  • Workflow repair – a helper layer to fix broken task flows without restarting everything.
  • Multi-channel runtime – works via CLI, terminal, and messaging platforms.
  • Local Graph UI – inspect memory and task graphs visually.

fagent is open-source and available here: https://github.com/fresed05/fagent

I’d love your feedback on:

  • How useful graph-based memory is compared to vector-only memory.
  • How intuitive the workflow inspection UX feels.
  • Any real-world workflows where you’d try a memory-focused agent.

I’m excited to see how the community experiments with long-running agents. Let me know what you think, or if you have suggestions for improvements!

Thanks for reading! 🚀

Top comments (0)