Work starts at the first error
A new error opens an investigation on its own, so work starts before the first customer complains.
finding out at 11:40 from an annoyed customerCapture & fix · Error monitoring
Start at the first error,not the first complaint.
Connect the error tracker you already run. Tore groups errors by fingerprint, opens an investigation on its own, and tells customers who write in that a fix is on the way.
Tore opens soon. Join the waitlist and we'll email you the day you can start.
app.tore.ai/northwind/errors
Example data incoming errorsSentryDatadogGrafana
grouped by fingerprintthreshold: 50 in 10 min
fp 9c1ewatchingthreshold crossed
TypeError: reading 'rows' · summary.ts:42
0events
fp 4f70below threshold
TimeoutError: search took 8.2 s · /api/search
0events
fp 07aaignore list
ResizeObserver loop limit exceeded
0events
INV-3107 opened on its own
Claude Opus · high
from fp 9c1e · nobody filed a ticketreading summary.ts and release 4.18.0root cause: summary.ts:42, empty result not guardedchange prepared · waiting for your engineer
Claude Opus · highinbox · customers writing inTore AI replying
No customer has written in yet. The investigation is already running.
The reports page is blank for me. Is something down?
We know about this one. A fix is on the way, and I'll message you here as soon as it's live.
Reports won't load since this morning?
Yes, we're on it already. You'll hear from me when it's fixed.
3conversations linked to INV-3107. One update goes to all of them when it's fixed.
Stop hearing it from customers.Hear it from the error.
Your error tracker already knows something broke. Tore turns that signal into work, before the queue fills up with "is it down?"
One bad deploy, one investigation
Identical errors group by fingerprint, so ten thousand events are counted as one problem.
ten thousand alerts in a channelThe noise stays out
Thresholds, an ignore list and an always-escalate list decide what is worth acting on before anything acts.
muting the alert channel"We know" instead of "tell me more"
Customers who write in about a known issue hear straight away that a fix is on the way.
"can you send a screenshot?"From the first stack traceto the customer who noticed.
Your tracker
Use the error tracker you already run.
Connect Sentry, Rollbar, Bugsnag, Datadog or Grafana. Tore reads errors, it never changes your tracker, and you can disconnect it in one click. A failed test run can start the work too.
- Read-only access to the tracker you already pay for
- Choose which projects and environments it watches
- A red test run in TestRail and similar starts an investigation
Settings · Error sources
Example dataNorthwind · 5 sourcesread-only
Sentrynorthwind-web · production, staging
connectedDatadogapi-gateway · production
connectedGrafanasearch-latency alerts · production
connectedRollbarnorthwind-mobile · production
connectedTestRailnightly regression · failed runs
triggerFingerprints
Ten thousand events, one problem.
Every event carries a fingerprint. Identical errors are counted as one group, tied to the release that started them, so a bad deploy becomes one investigation instead of a flood.
- One group per problem, with a live event count
- Tied to the release and commit that started it
- New groups stand out from errors you already know
Errors · since release 4.18.0
Example datarelease 4.18.0 · commit 7d2e19adeployed 09:12 by Dan
09:0009:1209:24
10,412events
1fingerprint
1investigation
Rules and limits
Noise is filtered before anything acts.
A threshold decides what is worth picking up. Known noise sits on an ignore list. Anything near authentication, billing or personal data goes to a person instead. Hourly and daily limits, and one stop switch, keep it in check.
- Thresholds per project and environment
- An ignore list and an always-escalate list
- Hourly and daily limits on what starts on its own, plus a stop switch
Intake · rule check
Example datanew fingerprints · last 10 minrule matched → outcome
ResizeObserver loop limit412 eventsignore listignoredChunkLoadErrorstaging · 26 eventsstaging: log onlyloggedTimeoutError: search37 eventsbelow 50 in 10 minwatchingSessionRefreshErrorauth/refresh.ts · 64 eventsalways escalate: authto a personTypeError: reading 'rows'summary.ts:42 · 1,284 eventsthreshold crossedinvestigatingstarted on its own: 3 of 10 this hourstop switch · running
Known issues
Customers hear "we know" straight away.
Once an error group has an investigation, Tore tells anyone who writes in about it that the team is on it. Their report is linked to the matching error event, and every conversation about it gets one update when the fix is live.
- "We know, a fix is on the way" instead of "can you tell me more?"
- The customer's report and the stack trace meet automatically
- Forty reports, one reply, sent once
INV-3107 · conversations
Example data
Hi, my weekly report is empty. Did I break something?
Not at all, it's on our side and we already know about it. A fix is on the way and I'll tell you here when it's live.
- 6 conversations linked to INV-310710:02
- Fix approved by your engineer11:48
- One update sent to all 6: it's fixed12:05
Tune it
Severity threshold High and aboveIgnore list Script error, bot trafficGroup errors Gemini Flash
5error trackers it reads: Sentry, Rollbar, Bugsnag, Datadog, Grafana
3lists you control: threshold, ignore, always escalate
1 clickto disconnect your error tracker
1 switchstops everything that started on its own
Keep exploring.
Let the first errorstart the work.
Connect the tracker you already run. The next bad deploy becomes one investigation, and the customers who notice hear "we know" straight away.