Notes from the cab
Architecture notes on investigation and remediation. No customer case studies, no invented metrics — just how the watch is supposed to run. Questions to office@alertnest.skin.
Note · the thread
The expensive part of a page is not the first alert. It is the second channel, the paste into a doc, the “what did we already try” that nobody can find. Alertnest treats the Slack thread as the console: intake, queries, charts, scripts, and the approve button all stay there.
Follow-ups inherit the investigation. Drop a screenshot or a log file in the same thread. The agent does not ask you to start over in another window.
Note · PII
Production logs are full of tokens, emails, and customer identifiers. The sandbox detects and redacts that material before anything is sent to the LLM. Credentials never sit in the agent’s filesystem — a proxy injects them at request time.
We do not use your data to train models for other customers. On-prem and VPC installs keep the glass on your side of the ridge. See security for the architecture drawing.
Note · write gate
One-click remediation is still a click. Restarting pods, rolling back configs, and rotating secrets are write actions. They require a human until you explicitly enable auto-mitigation for a named runbook.
Every action is audited. A rollback script sits beside the forward script. You keep the night watch — the agent keeps the map.