I built this after a training run crashed at 3am and I had no idea until morning. Every existing tool either blocks the main thread, requires a cloud account, or has no visual UI.

PulseLog is different in one way: log.info() costs 2.7µs and a browser tab opens automatically showing your logs in real time.

  pip install pulselog

Three lines to get started:

  from pulselog import Logger
  log = Logger("my-app")
  log.info("training started", epoch=1, loss=0.84)

What it does: - Non-blocking queue — O(1) calls, never blocks your thread - Live browser dashboard over WebSocket, zero config - Checkpoint system for tracking training progress (saves to SQLite) - Works offline, 100% local — no account, no data leaving your machine - 191 automated tests, benchmarked at 263k records/sec

Benchmarked on Python 3.12, Windows. Sustained throughput 290k/sec over 5 seconds. Zero dropped records under 8-thread load after v0.1.2.

I'm an indie developer. Would love brutal feedback on the API, the dashboard UX, and whether the checkpoint system is actually useful for ML workflows, data engineering pipeline or a simple python application.