Launching today
Runlog
Lightweight ML monitor — metrics, terminal logs, teams.
3 followers
Lightweight ML monitor — metrics, terminal logs, teams.
3 followers
Lightweight ML training monitor built for unreliable connections. Metrics and terminal logs stream live. Lose connection mid-run — nothing is lost, data buffers locally and syncs in order. Start fully offline, come online mid-training — run gets created on the dashboard and streaming continues. 3 lines to connect any PyTorch, HF, or Keras script. Currently in beta.
Hey PH! 👋 I'm Govind, solo builder of Runlog.
I spent months frustrated with existing tools — not because they lacked features, but because they cost a fortune, assumed perfect infrastructure, and still managed to lose my logs when my connection dropped at 2am.
So I built Runlog from scratch. No shortcuts, no wrappers around existing tools — every component designed for reliability and simplicity from the ground up.
The result: a training monitor that does more than W&B or Neptune for a fraction of the cost, and actually works in the real world where connections drop, machines crash, and teams need to collaborate without enterprise budgets.
What that means in practice:
Metrics and terminal logs stream live — print() statements are first-class citizens, not an afterthought
Fully offline-first — start offline, lose connection mid-run, come back online — nothing is ever lost, everything syncs in order automatically
Crash and dead run detection with instant alerts
Pause your training remotely from the dashboard without killing the process
Team workspaces with RBAC, run comparison, built-in chat, and shareable run links
3 lines to connect — PyTorch, HuggingFace, Keras, XGBoost, anything Python
Built lean, priced lean. No VC money, no inflated pricing to justify it.
Currently in beta with 25 open spots — every beta user gets full access to our top plan completely free for the entire beta period. No credit card, no trial limits. Just the full product. Happy to answer anything.