Skip to main content

Automated Pattern Detection

Continuously analyse your memory store to surface how well your agents are actually working together — where they contradict each other, where knowledge is going stale, and where one agent has become a single point of failure. Run a scan from Patterns in the dashboard, or through the API. A scan reads your entries, branches, pull requests, and outcome history, and returns a report grouped by category and severity.

Pattern Types

The detector looks for eight patterns across three categories.

Collaboration Health

Problems in how agents share knowledge.

Collaboration Insights

Opportunities rather than faults.

Reliability

Every detected pattern carries a severity of info, warning, or critical, the entity path it was found on, and a hint describing the suggested fix.

Tuning a Scan

The thresholds that decide when something counts as a pattern are configurable:

Alerting

Alert rules fire when a scan turns up something matching. Each rule filters by pattern type, minimum severity, and entity path, and has a cooldown so a persistent problem does not alert every scan. Alerts are delivered to a callback you register, which means routing is up to you — send them wherever your team already handles alerts.
Rules match on the pattern type strings above, such as knowledge_conflict or recurring_failure, and severities are lowercase: info, warning, critical.