Paste logs or run a sample. Timestamps are extracted where present; the rest keeps pasted order.
Investigating with the agent
Preview
No events loaded yet.
Timeline
Every event in order. Click evidence chips in the report to jump here.
No investigation yet. Run one first.
No report yet. Run an investigation first.
Summary
Observations
Hypotheses
Unknowns and limitations
Recommended next checks
No ticket yet. Run an investigation first.
Investigation steps
Ticket summary
Samples
Two ready-made cases. The second one shows the agent saying it does not know.
History
Kept only in this browser. Cleared when you clear site data.
Guide
How the agent works (30 seconds)
The agent is not a chatbot. The app first sorts your lines and flags patterns with plain code (no AI): error bursts, deploys near later errors, health shifts, repeating error messages.
Then the AI agent takes over with four read-only tools: search events, compare a deploy window, inspect health, find related events. It picks a check, the app runs it and returns the numbers, the agent reads them and picks the next check (at most 6). Finally it writes hypotheses tied to event IDs, lists what is still unknown, and suggests concrete fixes for a human to apply. It never restarts, rolls back, edits, or deletes anything.
Limits: 200 lines or events per run. Timestamps are extracted where present; the rest keeps pasted order. Do not paste secrets or customer data.