The moment an issue opens, your Aivell reads it, labels it, hunts for duplicates and gives a useful first response — asking for the version and steps a bug report is missing. On Fridays it sweeps the backlog. It comments and labels; closing decisions beyond obvious duplicates stay human.
starts when: A new issue opens on one of your repositories
Issue triage is a tax paid in context switches. A new issue lands mid-function: is it a duplicate of #482, a question the docs answer, or a real bug missing its version number? Working that out takes ten minutes of archaeology, so it gets postponed — and an issue tracker where nothing gets a first response for a week quietly tells users to stop reporting things.
The backlog compounds. Unlabeled issues can’t be filtered, near-duplicates breed independent comment threads, and the eventual cleanup weekend recurs like a bad habit. None of this is engineering; all of it steals engineering time.
Reporters get answered while the report is still warm, maintainers get issues that arrive pre-sorted with the missing details already chased — and the triage rules stay a paragraph of plain language you can change over coffee.
No flowcharts, no code — a chore is just a message to your Aivell, in your own words. It shapes it into a solid, guarded procedure and follows it to the letter. This one:
when an issue opens, read it fully, including pasted logs
label it — bug, feature, question or support
hunt for duplicates; link the original and close the copy politely
questions the docs already answer get a first reply with the link
bug reports missing version or steps? ask for them, friendly, once
security reports get no public reply — mail Chiara immediately
never close a bug or feature request; humans decide those
friday, list issues silent for 60 days and label the unlabeled
Connected with single-click passkeys — no copied tokens, no OAuth hell.
Every tool can be limited to exactly what this chore needs. For this one, you might set:
It comes to you. The box fetches the issue, and all the reading and reasoning happen onboard — the only things that travel back to GitHub are the label and the comment. No second AI service ever sees your tracker.
It closes exactly one thing — duplicates, with a link to the original — because the guardrail allows nothing more. Every other issue stays open for a human, and the tone of first replies is whatever you wrote in the chore.
Silence in public, speed in private. The chore forbids a public reply and mails the maintainer directly instead — the one triage case where a fast templated response is exactly wrong.
No. A guardrail restricts it to issues and comments — no pushes, no merges. Triage is a librarian's job, and the Aivell is kept firmly at the librarian's desk.
Any system that can send a webhook can trigger a chore — your shop, your form tool, your monitoring. The glue logic runs on a box in your office.
Password questions and how-do-I requests answered from your own runbooks; real incidents structured and escalated. All behind your firewall.
Merge CSVs, convert files, dedupe exports — your Aivell writes and runs the script in a no-network sandbox on the box. Data in, result out, nothing else.
An Aivell is an on-premise AI colleague: a small box that plugs into your network, runs its own AI onboard and takes the repetitive work off your desk. Prompts, documents, data — nothing ever leaves your office. Unplug the internet: it keeps working.
You describe each task as a chore, in plain language. Your Aivell turns it into a solid, guarded procedure and runs it in the background — with guardrails on every tool, a complete audit trail, and approvals in your hand when you want them. One-time setup, fixed monthly fee. No tokens, no overages.
onboard inference · air-gapped · fixed monthly fee