Changelog
What's new
A running log of what we shipped. Newest first.
v1.4
Grafana Loki source, evidence deep links, and Slack delivery GA
- Added Grafana Loki as a first-class telemetry source. Log-based diagnoses now pull from Loki alongside Datadog and Sentry with the same ingest quality.
- Evidence citations now link directly to the source log line or trace span. You can verify a quoted line in one click without searching the raw log.
- Slack delivery of the ranked cause, evidence summary, and proposed fix is now generally available for Team and Scale plans. The result posts into your incident channel within seconds of the diagnosis completing.
- Fixed an edge case where a deploy entry with no author field caused the ingest step to stall without surfacing an error.
v1.3
Router default updated, diagnosis latency down roughly 22%
- Cross-signal reasoning is now routed to a frontier model by default. Median end-to-end diagnosis latency is down roughly 22% compared with the previous routing table.
- Fast triage steps, including signal classification and context pruning, now use a lighter model. This frees headroom for the reasoning layer without reducing output quality.
- The model router now retries the next candidate in the registry on a provider timeout, so a single lab outage no longer blocks a diagnosis mid-alert.
- Evidence quotes are now ranked by how directly they support each root cause rather than by log timestamp. The most diagnostic lines surface first.
v1.2
Scheduled maintenance complete, diagnosis queue introduced
- Planned maintenance on the diagnosis engine completed on 2026-06-05. Total window: 9 minutes. All diagnoses queued during the window replayed automatically with no data loss.
- Introduced a durable diagnosis queue so that brief engine downtime no longer drops an in-flight request at the moment of an alert.
- Improved error surfacing when a connection credential expires. The connector now names the specific source that failed rather than returning a generic timeout.
v1.1
GitHub deploy correlation, past-incident matching, and per-cause confidence
- Ketl now correlates alert time with recent GitHub releases and pull requests. If a deploy landed within the look-back window, the diagnosis cites the specific commit, author, and changed files.
- Past-incident matching searches your incident history for similar cause signatures and attaches matched incidents as supporting evidence, so prior knowledge is not relearned during an active page.
- Confidence scores now appear on every candidate cause rather than only on the top-ranked result. Thin evidence lowers confidence rather than inflating certainty on secondary causes.
- Reduced the minimum look-back window for traces from 15 minutes to 5 minutes, capturing failures in services with fast error recovery.
Stay on the current version
New integrations and engine improvements ship every few weeks. Start free to get access as we ship.