Trace your AI sessions and training runs in one place.
claude-monitor is a wandb-style dashboard for the work your agents and models actually do: trace waterfalls for Claude Code & Codex sessions, live metric charts for training runs, and one-click sharing — with HF dataset and model links built in.
Trace waterfalls
Every tool call, message, and subagent rendered as a navigable timeline with diffs and syntax highlighting.
Training runs
wandb-style metric charts with EMA smoothing, log scale, and resizable tiles. Scalars and histograms.
Share anything
Publish a run, a project, or a dataset behind a private link — or start sending with no account at all.
Start in 30 seconds
Install the SDK and log a run. No API key required to try it — you'll get a public link and can sign in to claim your data later.
shellbash
pip install claude-monitor-sdk
train.pypython
import claude_monitor as cm
# No API key needed to start — your run gets a public link.
run = cm.init_run(project="demo", name="my-finetune",
config={"lr": 1e-4, "batch": 32})
for step in range(1000):
run.log({"train/loss": loss}, step=step)
run.link_model("you/my-finetune") # link the final HF model
run.finish() # then log in to claim it