AI based alert thresholds

[Vadman97]

embedding based anomaly detection?

[linear]

HIG-4635 AI based alert thresholds

[greptile-apps]

To implement AI-based alert thresholds with embedding-based anomaly detection, you'll need to integrate the embeddings functionality from /backend/embeddings/embeddings.go with the alerting mechanisms in /backend/alerts. Specifically:

1. Use GetEmbeddings from embeddings.go to generate embeddings for error objects or sessions that you want to monitor for anomalies.
2. Develop a new anomaly detection function in embeddings.go that utilizes these embeddings to identify outliers based on your specific threshold criteria.
3. In alerts.go, add a new alert type for embedding-based anomalies. This involves creating a function to trigger alerts when anomalies are detected.
4. Utilize the routing.go and sessionalerts.go for generating URLs and session alert objects related to the detected anomalies for notification purposes.
5. For integration with external systems (Discord, Microsoft Teams, or webhooks), extend the functionalities in the respective directories under /backend/alerts/integrations to handle the new type of alerts.

This approach leverages existing structures for embeddings and alerts, focusing on the integration and expansion of these functionalities to support embedding-based anomaly detection.

References

/backend/alerts
/backend/embeddings

Ask Greptile