Skip to content
HiveGuard

hiveguard ml

HiveGuard uses a lightweight ML model to score challenge responses. The model learns from accumulated labeled data to detect bots more accurately over time.

ML commands require dashboard credentials.

ml status

Check whether the model is trained and ready.

Terminal window
hiveguard ml status

Output:

Model ready:             yes
Sessions with features:  3,241
Last trained:            2024-01-01T10:00:00Z

If the model hasn’t been trained yet:

Model ready:             no
Sessions with features:  142
Last trained:            never


Run 'hiveguard ml train' once you have enough labeled sessions.

ml train

Trigger a training run.

Terminal window
hiveguard ml train

Training runs asynchronously on the server. The command returns immediately after submitting the job:

Training job submitted.
Check status with: hiveguard ml status

Training requires a minimum number of labeled sessions. If there isn’t enough data:

Error: not enough labeled sessions to train (need at least 100, have 42)

When to retrain

Retrain the model when:

  • You’ve uploaded a significant number of new items (> 500)
  • The solve rate changes unexpectedly
  • You’ve added a new modality to your dataset