Analyze exceptions with AI
Start an AI investigation from an exception, review whether the evidence supports code changes, and share the findings with a coding agent.
Start an AI investigation from an exception’s detail panel to inspect the failure in its service and time window. Last9 uses the exception details to gather related telemetry, then reports whether the evidence supports starting code changes or needs more investigation.
Start from an exception
- Open Exceptions in Last9 and select the time range you want to investigate.
- Click an exception row to open its detail panel.
- Click Investigate with AI. The investigation starts when the dialog opens.
- Review the streamed investigation notes and the resulting analysis.
The action is available where exception AI investigation is enabled. If it is disabled, ask an organization admin to check Enable Last9 AI in AI Settings.
What the investigation uses
You do not need to retype the exception details. Last9 supplies the selected exception’s identity and fingerprint, service, operation, cluster, and time window.
Last9 uses the selected exception details to look for supporting evidence:
| Evidence | What it helps explain |
|---|---|
| Exception details | The failure being investigated and its recorded context. |
| Failing traces | The operations and dependencies involved in affected requests. |
| Correlated logs | Error messages and request details related to the failure. |
| Service dependencies | Where an upstream or downstream service may be involved. |
| Change events | Changes near the incident that may warrant investigation. |
The available evidence depends on what is recorded for the selected exception and time window. For example, if the exception cannot be linked to a trace, the investigation may be unable to confirm a cause in a downstream service.
Review the result
The result includes a written analysis and a status indicating whether the findings support starting code changes:
| Status | What to do next |
|---|---|
| Ready for code changes | Review the supporting evidence and source context, then hand the investigation to the engineer or coding agent that will prepare the change. |
| More evidence needed | Review the listed gaps. Collect the missing source, reproduction details, or telemetry before treating a proposed cause as established. |
The status describes whether the investigation is ready for an engineer or coding agent to prepare a change. It does not mean a fix has been written, tested, or deployed.
The investigation is your own and has no share link. Use Copy investigation to pass the report to a teammate or a coding agent.
Use Run again to rerun the investigation. Use Continue in AI Assistant to follow up on the findings. For example:
Which missing evidence would distinguish the two possible causes?Check whether the related logs or traces can resolve the uncertaintyin the same incident window.To investigate a different period, return to Exceptions and select the new time range. Start another investigation from the relevant exception.
Hand the investigation to a coding agent
Click Copy investigation to copy the report. It includes the exception details, findings, evidence, possible causes, source code information, reproduction details, and remaining gaps.
Paste the report into your coding agent in the relevant repository so the agent can locate the affected code and check a proposed fix. When the investigation is ready, the agent can use the report to prepare a draft pull request. Code review, testing, merge, and deployment remain separate steps.
The exception workflow investigates the failure and prepares a report. It does not create a pull request or deploy a fix from the exception panel.
For a broader incident, refer to Investigate an incident with AI. To connect telemetry tools to the coding agent that receives the report, see Last9 MCP setup.
Troubleshooting
-
Investigate with AI is not available in the detail panel
Ask an organization admin to check Enable Last9 AI in AI Settings.
-
The status reads More evidence needed
Review the listed gaps. Collect the missing source, reproduction details, or telemetry before you treat a proposed cause as established.
-
The investigation cannot confirm a cause in a downstream service
Check whether the exception is linked to a trace in the selected window. Select a window where the failing requests are recorded, then start the investigation again.
-
The analysis covers the wrong period
Return to Exceptions, select the new time range, and start another investigation from the relevant exception.
-
The copied report lacks the code context your agent needs
Open the evidence links and add the missing source or reproduction details before you paste the report into the coding agent.
Please get in touch with us on Discord or Email if you have any questions.