A small hobby project of mine: JuPipe. A file-based, multi-step LLM pipeline runner.
For anyone who wants to chain LLM calls — without framework overhead, GUI or vendor lock-in. The idea behind it: you simply define your workflow in plain text files. That’s it.
Tools like n8n or Windmill can do something similar, and much more — but they come with significant overhead. I wanted something minimal, with no server and no GUI: just text files and a CLI.
What it does:
- Chains LLM calls: each step can access the results of previous steps
- Any LLM provider via LiteLLM
- Loops, conditions, inline Python, JSON capabilities
- TXT or PDF input
Supports both local models and cloud APIs — which mattered to me, because I did not want to depend on a single cloud service.
Just a hobby, no support, no promises. But maybe useful to someone. Feel free to share: github.com/juwy/jupipe
JuPipe shows that you do not necessarily need a large framework to build LLM pipelines — plain text files and a CLI are enough for many use cases, and you stay independent of any single provider.