symboulionAll posts · 12 Oct 2026
Why a shared table has to be cheap
The goal is that server and AI for one cluster cost less than 200 euros a year. Here is how the app is built around that goal.
Clusters work with small budgets. A tool that costs more than one company's yearly membership fee will not survive the first cost review.
So a goal was set from the start: server and AI for one cluster under 200 euros a year. That is a goal, not a measured figure. The pilot measures the real cost from day one and people judge at the end whether the goal was met.
- Checks messagesA fast and cheap model checks messages against the rules of the table.
- DraftsMinutes and monthly reports, which people confirm.
- Remembers what mattersOnly topics, decisions, minutes and reports are processed for search by meaning, not every message.
- No audioThe meeting recap comes from the tool the cluster already uses.
- Watches the budgetMonthly spend is tracked per organization, with a warning at 80 percent of the budget.
For every AI call only a short record is kept: which version of the model and the rules was used. The full context is frozen only when someone presses This could be better.
The model is chosen separately for each function. When a cheaper model that works just as well appears, one setting changes, not the code.
Pilot measurement tracks nothing extra about people, no page visits and no reading time. Only things that already exist are counted: messages, tasks, minutes, votes and cost.