If you often do things like:
- tweak a few parameters and rerun the same script over and over
- keep shell commands in terminal history and lose track of what actually worked
- run multiple experiments at once and end up with mixed, unreadable logs
- want lightweight experiment management without pushing your workflow into a heavyweight cloud platform
then Pyruns might be worth a look.
What it is
Pyruns is a local-first experiment workspace for Python scripts and shell tasks.
It is built around a simple idea:
experiment tooling should stay close to the way people already work
That means:
- keep using your own scripts
- keep using your own terminal and conda environment
- keep everything local
- keep task configs, logs, notes, and run history on disk in a predictable workspace
What it does
- visual parameter editing for Python script workflows
- batch task generation in form mode
- YAML-based single-task editing
- shell task management with host-terminal semantics
- task search, filtering, rerun, delete, pin, and notes
- terminal-style live logs in the browser
- CSV export for recorded metrics
- one isolated config/log directory per task
The main entrypoints
The primary path is script-first:
pyr train.py
pyr train.py config.yaml
This is the recommended first experience for most users.
There is also a shell-first path:
pyr
That opens a shell workspace for the current directory.
It is useful when what you want to manage is a set of commands rather than a single Python script.
Why it feels different
Pyruns is not trying to replace your workflow with a platform.
It feels more like a practical local workbench for people who already have:
- Python scripts
- YAML configs
- shell commands
- local logs
- repeated experiment runs
and simply want those pieces to become easier to generate, run, inspect, and revisit.
Current stack
- React frontend
- FastAPI backend
- Home / Generator / Manager / Monitor UI flow
Where it fits well
-
argparse-based training scripts -
pyruns.load()/ YAML-driven workflows - local ML or automation experiments
- shell-heavy command workflows
- users who want local visibility without cloud lock-in
Links
Install
pip install pyruns
Start
pyr train.py
Top comments (0)