What we were handed.
Every tool grows a dashboard, and nobody opens dashboards. Our team lives in Discord — so every status check or routine task meant a context switch away from where the conversation was happening.
Meanwhile the knowledge of how to do those tasks lived in people's heads, which doesn't scale and doesn't survive a vacation.
The system.
An AI assistant that runs as a Discord bot, backed by our self-hosted model stack. It can hold a conversation, but the point is that it acts: it calls real tools — status checks, queries, jobs — instead of just describing them.
Its skills live in a library of written playbooks the bot can list and follow. Teaching it a new job means writing a short document, not shipping new code.
A playbook library the assistant reads and follows — knowledge as files, not folklore.
Real actions wired in: checks, queries, and jobs, not just chat.
Hybrid local/cloud models — routine work stays on our hardware.
It lives in Discord because that's where the team already is.
What changed.
Answers and operations happen in the channel where the question was asked. And because skills are documents, tribal knowledge steadily turns into things the assistant can do for anyone on the team.
The next study could be yours.
Bring us the problem — the messy, manual, “there has to be a better way” kind. We'll design the system, build it, and run it.