Runlog

Runlog is an AI experiment tracking tool that monitors and controls machine learning training runs in real time for ML engineers and research teams. Built for unreliable connections, it streams metrics and terminal logs live, and its offline-first design buffers data locally so nothing is lost if the connection drops mid-run — runs even start fully offline and sync in order once you reconnect. You connect any PyTorch, HuggingFace, Keras, or XGBoost script in about three lines of code. Beyond passive logging, Runlog adds live control: adjust hyperparameters mid-run with "live knobs," pause or stop training from the dashboard without restarting, and rely on automatic crash detection for exceptions, OOM errors, and silent hangs. Metric charts are auto-detected, and you can log custom key-value pairs. For teams, it offers workspaces with role-based access control, cross-user run comparison, run annotations anchored to training steps, built-in group chat, and public read-only shareable links. Conditional email alerts fire when metrics cross thresholds, and a checkpoint ledger keeps metric snapshots. Runlog is currently in beta.

ML engineers and research teams use Runlog to track, monitor, and control PyTorch, HuggingFace, Keras, and XGBoost training runs in real time, even over unreliable connections.