Paste the error.
Get the story.
A read-only investigation agent for small teams. It rebuilds the timeline from your logs, checks it with an AI agent, and shows what happened, what broke, and what to check next, every claim linked to evidence.
The agent, concretely
Not a chatbot. A loop: the model picks a read-only check, trusted code runs it, the model reads the numbers and picks again.
Filter by service, severity, signature, time window.
Errors and health before vs after a deploy, with a no-causation caution.
Min, max, average over any window you name.
Same-service events near any event.
Budgets per run: 5 model calls, 6 tool calls, 1500 output tokens. Unknown tools and anything resembling remediation are rejected.
Same failure 6 times: CheckoutError, payment gateway timeout
Unknowns stated too: deploy diff absent, infra metrics absent. When evidence runs out, the report says so.
Show me why
Every claim carries its evidence.
Facts, hypotheses, and unknowns look different from each other, so you never have to parse a wall of prose under pressure. Click any evidence chip and the timeline jumps to that exact event and highlights it.
Built on AWS, kept small
Your next incident, reconstructed.
Open the dashboard, paste the logs, read the report. About a minute, end to end.