Nothing ages out: the case against expiring history

Every chat tool sells you the same thing twice. First it sells you a place to make decisions. Then, once you have made a few thousand of them, it sells you access to what you decided.

The 90-day cliff is not a technical constraint. Storing text is close to free, and the index that makes it searchable is the same index the tool already runs. What expires is not the data — it is your permission to read it.

What you actually lose

The obvious loss is the decision itself: the thread where you agreed the discount, the message where someone flagged the risk, the reply that explained why the schema looks like that. Those are annoying to lose and easy to picture.

The expensive loss is quieter. When history is unreliable, people stop treating chat as a record. They start re-asking instead of searching, because searching has burned them before. Within a quarter you have a team that documents nothing and repeats everything, and the tool that caused it looks blameless.

Why an AI layer makes it worse

An assistant bolted onto a chat app with a retention limit inherits the limit. Ask it what your team decided last quarter and it will answer confidently from whatever fragment it can still see. That is worse than a blank — it is a plausible answer with no way to check it.

Memory is the substrate, not the feature. An agent that can only read the last twelve weeks of your company is an agent with amnesia and good manners.

What we did instead

Huub keeps everything, from the first message, on every plan including the free one. The free plan caps the number of people, not the history, because the number of people is the thing that actually costs us money.

Search reaches the first message the workspace ever sent, and it reaches inside your files while it is there. When the agent answers a question, it cites the channel, document or record it read, so you can go and check.

None of this is generous. It is just what a record is supposed to do.