New Error Detection: An Issue When a New Kind of Error Appears

Last verified on Nov 8, 2026.

What it catches

Novel event fires when a new kind of error appears in your logs. It runs on every production project. The usual cause is a change: a deploy ships a code path that fails in a way nothing in your system has failed before, a dependency upgrade rewrites an error message, or a new service logs its first failure.

You do not write a rule first. There is no query to define and no threshold to pick. When the detection fires, Sazabi opens an issue in your project and you take it from there.

Example: a deploy adds a new failure

This is an example, not output from a real project. You deploy the drafts-api service at 14

2026-11-08T14:03:41Z ERROR drafts-api save_draft failed: skipping save, draft locked by session 8f21 (user_id=4412)

Nothing like it was in your logs before the deploy. Sazabi opens an issue. An example of what that issue could say:

• Name: Draft saves failing with "draft locked" in drafts-api
• Description: A new error began at 14
  UTC, right after a deploy of drafts-api, and the same message is repeating for several users.
• Severity: medium

Your on-call engineer reads that in Slack or the issue inbox and starts from an analysis instead of a raw log line.

What you see in Sazabi

The result is an issue in the project's Issues inbox. According to the Sazabi docs, an issue carries a stable name, a short description, a severity of low, medium, high or critical, an optional linked component and a status. It arrives already investigated: the agent attaches what happened, why and the evidence behind it.

From the issue you can:

• Open its thread and ask the agent follow-up questions with the issue context already loaded.
• Hand the fix to a coding agent such as Claude Code, Codex, Cursor or Devin. The sazabi issues agent-prompt command prints the prompt.
• Resolve it, ignore it or mute it from the dashboard, from Slack or Microsoft Teams, or from the CLI.

If the same problem comes back while an open or ignored issue exists, the recurrence is folded into that issue. A recurrence of a resolved issue opens a new one.

The CLI can also list the underlying signals. sazabi signals list --type new_error shows what the detector reported, and each one carries the error log line, its service and its severity.

How you get told

A new issue produces an Issue triggered notification. By default it lands in your in-app bell feed and skips email, and you can turn on email, Slack DM or SMS under your own notification settings. Shared destinations work differently: a Slack channel, Microsoft Teams channel, webhook, PagerDuty, incident.io or Rootly only receives the event when a delivery rule names it. Rules can match on severity, component or owning team, so a payments team can get only the high and critical issues for its own services. See Slack and PagerDuty for the channel setup.

What you need

You need logs from a production project. The failing service has to send its error logs to Sazabi, and any supported source works: a connected platform such as Vercel, an OpenTelemetry exporter, Fluent Bit or Vector on your hosts, or an agent you already run such as the Datadog Agent in dual-shipping mode. The integrations list has the rest.

Confirm the data is flowing on the Intake page. A source that sits at zero when you expect traffic is an ingestion problem, and detection cannot see what never arrives.

What it does not cover

Novel event looks for a new kind of error and nothing else. It does not cover:

• An error your logs already contain, even when it gets worse.
• Something that should happen and stops happening.
• Problems that never write an error to your logs.

Sazabi runs other detections for some of these. A jump in how often an event is logged, an event that disappears and a source that goes silent are described on the volume shifts, vanished events and service went quiet pages, and the reliability issues page lists everything that feeds reliability issues. You can also ask the agent about anything else in a thread, and an investigation you start can end with an issue.

Issues Sazabi detects are metered, so check pricing. For the vendor-neutral version of this topic, read the guide to detecting new errors. Novel event is also one of the ways Sazabi catches a bad deploy, which the regressions page covers.