- +northline_support_runworkflow
- +triage_agentagent
- +lookup_customertool
- +gpt-5.5-2026-04-24llm
- +knowledge_agentagent
- +search_kbtool
- +resolution_agentagent
- +create_refundtoolPII in Outputs
neatlogs investigates.
You review the fix
neatlogs gathers the evidence, finds what caused the failure, and prepares a fix brief your coding agent can act on. You review it like a pull request.
Findings go where your team already works:
SlackLinearJiraGitHub
Claude Code
CursorMCPEvidence inSlack conversationcustomer impact and context
Production tracesruns · spans · tool calls
Slack conversationcustomer impact and contextIssue contextLinear · Jira · support
Change historydeploys · commits · config
Cause isolatedEvidence agrees
Fix brief ready forClaude Codeact on the briefCursorprepare the changecause · evidence · next steps
Claude Codeact on the briefCursorprepare the changeGitHubreview the pull request
Every issue starts with
the evidence attached.
However an issue appears, neatlogs connects it to the production evidence before the investigation begins.
A detection fired
A condition, pattern, PII check, or classifier flagged the exact span where the behavior appeared.
northline_support_turntrace · 4a365745
Is this refund valid under the customer’s returns policy?
An eval failed
A person or AI evaluator judged an output against your standard, and it did not clear it.
NeatlogsSearch workspace
@neatlogs refunds are failing again for Northline. Can you check the latest run?
I found the matching production run and attached the reported outcome.
Matched production evidencenorthline_support_run
Linkedrun_1842 · create_refund · 503Someone reported it
A customer or teammate flagged the outcome in Slack, Discord, email, or Intercom. neatlogs connects their words to the run they describe.
nProduction monitor
Failed operation
create_refund returned 503
Retry budget exhausted after three attempts.
Request sentPayments API
Three retriesExponential backoff
Execution failedService unavailable
neatlogs noticed
It found a failed tool call, broken execution, or suspicious behavior without waiting for an external trigger.
Evidence becomes a case
Now neatlogs
works the case.
Most tools give you a failed trace and somewhere to start reading. neatlogs investigates across everything already known about the issue.
northline_support_runrun_1842AIrefund_agentissue_refundPolicyViolation
northline_support_runrun_1848AIrefund_agentissue_refundPolicyViolation
EvaluationRefund policy review
100%4 items
One timeAI
Slack · #support-opsNowAO
Amara Osei
Refunds are failing again for Northline. The latest customer was outside the return window.
RM
Ravi Menon
I found the same outcome on another enterprise account. Both runs skipped escalation.
You get one report.
With 4 things your team needs.
n
Refund policy investigationINV-1842 · final report
InvestigatedRoot cause
ConfirmedEvidence and reasoning
12 linkedRecommended action
ReadyVerification
3 checksReview it. Ship it. neatlogs sees if it held
Bring the investigation to your team, hand a bounded fix brief to your coding agent, and keep watching the same failure path after the change ships.
Root causeEscalation check was skipped
The refund path continued after the policy lookup returned an exception.
@neatlogs this account is in the enterprise exception group. Did the run load that policy?
Good catch. I found a second policy lookup and added it to the evidence.
Discuss the evidence
Comment on the report, challenge the reasoning, and bring in the person who knows the full context.
SUP-482High severity
Require escalation before enterprise refunds
Prevent `create_refund` when the account policy requires a specialist review.
- Evidence
- 4 spans + report
- Acceptance
- 3 verification cases
Create the ticket
Send the issue to Linear or Jira with the report linked and its severity, evidence, and acceptance criteria intact.
Claude CodeMCP briefFix brief
Gate refunds behind policy escalation
Change only the decision path used by `resolution_agent`.
- Root causeException result ignored
- Relevant spanslookup_policy → create_refund
- ExpectedEscalate before tool call
- Verify3 supplied regression cases
Send it to your coding agent
Give Claude Code, Cursor, or another MCP-compatible agent the root cause, relevant spans, expected behavior, and verification cases instead of an entire trace dump.
Customer replied“The account was upgraded yesterday.”
Analysis rerunEvidence changed the caseComplete
Previous
Policy exception was missing.
UpdatedStale account tier caused the wrong policy to load.
Investigate again
Re-run the analysis when a customer replies, a teammate adds context, or new evidence changes the case.
Fix deployed
Enterprise refund guard
Watching the same failure path in production.
Policy loaded248 runs
Escalation respected100%
No recurrence7 days
Watch the fix
neatlogs monitors your agent in production, and re-surfaces the issue if it returns.
Connect your current platform, choose a time range, and bring your production history into neatlogs. Pick a run that went wrong and see what the investigation finds.
Langfuse
LangSmith
Braintrust
Phoenix
Raindrop
OpenTelemetryDetections find. Evals judge.Investigations diagnose.
Each layer answers a different question about production behavior.
CapabilityThe question it answers
01Trace
What happened during this run?
02Detection
Where did a known condition or behavior appear?
03Investigation
Why did it happen, and what should change?
04Eval
Did the behavior meet an explicit standard?
Product questions and answers
The practical details about issues, investigation evidence, recommendations, and verifying the fix.
Selected question01 / 11
What’s an issue?
An issue is a flagged problem in an agent’s behavior. It carries a severity, a source, and the runs it belongs to. Issues come from a fired detection, a failed eval, a person reporting something, or neatlogs raising one on its own.
An issue is a flagged problem in an agent’s behavior. It carries a severity, a source, and the runs it belongs to. Issues come from a fired detection, a failed eval, a person reporting something, or neatlogs raising one on its own.