Skip to main content
An analysis is a standalone AI investigation session. Where incidents are tied to a specific event in your system, an analysis gives you a free-form way to investigate any scenario — with full control over which integrations are consulted and what context the agents receive.
Each analysis consumes one investigation from your monthly plan allowance. Check your usage at Dashboard > Billing.

​
Creating an analysis

1

Open the new analysis form

Go to Dashboard > Incidents and click New Analysis, or start one directly from any incident that needs deeper investigation.
2

Select a severity level

Choose the severity that best describes the issue you’re investigating: critical, high, medium, low, or info. This helps CauseFlow prioritize which agents to activate and how aggressively to search for root causes.
3

Choose integrations

Select which connected integrations the analysis should use. By default, all connected integrations are available.Limiting integrations to the relevant systems speeds up the investigation and reduces noise in findings. For example, if you’re investigating a database slowdown, select only your database and monitoring integrations.
4

Write a description

Describe what you’re investigating (10–4000 characters). Be specific:
  • What symptoms are you observing?
  • Which services are affected?
  • When did the problem start?
  • What changed recently (deploys, config updates, traffic spikes)?
The description is the primary instruction for the AI agents. More context leads to more accurate findings.
5

Submit

Click Start Analysis. CauseFlow dispatches the relevant agents immediately.

​
Analysis detail

Once an analysis is running, the detail page shows its progress in real time.

​
Investigation progress

A progress indicator shows which agents are queued, running, and complete. Each agent operates independently — you’ll see findings appear as individual agents finish rather than waiting for all of them. The page refreshes automatically while the investigation is in progress. You don’t need to reload to see new findings.

​
Agent findings

Each finding entry shows:
  • The agent that produced it (for example, logs agent, metrics agent, infrastructure agent, code agent)
  • A summary of what the agent discovered
  • Supporting evidence — excerpts, anomalies, and references to the specific data the agent analyzed

​
Root cause and recommendations

When all agents have reported, CauseFlow synthesizes a root cause analysis and a set of remediation recommendations. The root cause section explains the likely chain of events that caused the issue. The recommendations section proposes specific actions to resolve it.

​
Re-investigating with additional context

If the initial analysis is incomplete — for example, the root cause is unclear or an agent couldn’t access the relevant data — you can add more context and re-run the investigation. On the analysis detail page, click Add Context. Enter additional details that might help the agents focus:
  • “The problem started immediately after the deploy at 14:00 UTC.”
  • “Only users in the EU region are affected.”
  • “We ruled out the database — queries are responding normally.”
Submit the additional context to start a new investigation pass. The agents will incorporate your update alongside the previous findings.
Re-investigation is especially useful when the initial description was broad. Use the first run’s findings to identify the most promising area, then re-investigate with a focused prompt.