Skip to content
Hosted service in public preview. 3.0 installs from GitHub until it is on PyPI.

Dashboard

Terminal window
agenticml ui # http://127.0.0.1:6006
agenticml ui --open --port 6006 --host 127.0.0.1
agenticml ui --read-only

agenticml ui binds to 127.0.0.1 on port 6006 and reads straight from the local store. --dir picks a different store (otherwise AGENTICML_DIR, then ~/.agenticml).

Terminal window
agenticml ls # newest first
agenticml ls --project demo --json

ls prints the run id, project, name, status, created time, and last step.

A run stays running while its process holds the run’s lock, even if it logs rarely. If the process is gone, the run shows as crashed. On filesystems without locking, runs untouched for more than 120 seconds show as stale.

The dashboard pages do not show run notes. Launch notes (--notes / init(notes=)) are still stored with the run.

The server’s only write is the /api/runs/<id>/notes endpoint. --read-only disables it. Use it if you expose the dashboard on a non-loopback --host.

In addition to agenticml ls, the CLI provides commands to inspect details, export data, and update finished runs.

Terminal window
agenticml show RUN_ID
agenticml show RUN_ID --json
agenticml export RUN_ID -o metrics.csv
agenticml export RUN_ID --key loss --key accuracy --format json

agenticml show RUN_ID displays the run header (ID, project, name, status, task, created time, parent, and tags/group/job type/notes when present), configuration (hiding internal keys starting with _), summary, a metrics table (KEY, LAST, STEP, POINTS), an artifacts table (NAME, TYPE, VERSION, SIZE, PATH), and media item counts. Pass --json to output a strict JSON document; non-finite floats are output as null. LAST and STEP reflect the last written record (or the overview’s largest step and last finite value for series pruned by sync --prune). Missing runs, bad IDs, or corrupt metadata print an error on stderr and exit 1.

agenticml export RUN_ID dumps raw metric series. Default format is CSV (key,step,value in write order, float32 formatted exactly, non-finite values as nan, inf, -inf). --format json outputs {"run_id": ..., "series": {key: [[step, value], ...]}} with non-finite values as null. Filter series with repeatable --key K. Use -o PATH to write to a file. Unknown keys fail before writing (exit 1). Pruned series are skipped with a warning; if all chosen series were pruned, export exits 1.

Ended runs can be updated directly from the CLI. edit, artifact add, and finish refuse live runs whose lock is held (exit 1). They do not mark runs as running and do not refresh heartbeats. Changes made with edit, artifact add, and finish are uploaded by the next agenticml sync, even for runs that were already synced. Runs pruned by sync --prune cannot be edited.

Terminal window
agenticml edit RUN_ID --tag best --summary test_acc=0.913 --group sweep-3
agenticml artifact add RUN_ID model.pt --name checkpoint --copy
agenticml finish RUN_ID --status crashed
agenticml note RUN_ID --set "Validation score confirmed"
  • agenticml edit RUN_ID: Updates metadata on an ended run. --tag T and --untag T are repeatable (add if absent, remove every occurrence). --notes TEXT updates the launch note ("" clears it). --group NAME and --job-type NAME set the group and job type (cannot be cleared). --config KEY=VALUE and --summary KEY=VALUE merge values using the same JSON/string parsing rules as run --config. --unset-summary KEY removes a summary key. Project, name, task, parent, and run ID cannot be changed.
  • agenticml artifact add RUN_ID PATH: Records an artifact on an ended run (--name defaults to file stem, --type defaults to file, --metadata accepts a JSON object). --copy stores file bytes in the object store; otherwise only path, size, and mtime are recorded. Version increments per name.
  • agenticml finish RUN_ID --status finished|crashed: Finalizes a run left running, stale, or crashed (or toggles finished/crashed), writing overviews and finished_at while preserving summary metrics. If the run is already in that status, exits 0.
  • agenticml note RUN_ID: Reads or updates the run’s local Markdown note (runs/<id>/notes.md) using --set TEXT, --file PATH (- for stdin), or --clear. The notepad is local and not synced. Unlike the other commands, note works even while the run is live. Without flags, it prints the current note.

Note that agenticml note modifies the local note file (notes.md). To change the run’s synced launch note, use agenticml edit --notes TEXT.