Cost and model selection
Where tokens actually go.
What costs nothing
Search. Embedding runs locally on your own hardware, so indexing and querying cost no tokens regardless of corpus size.
That is most of the traffic in a normal instance.
What costs tokens
| Operation | When it runs | Rough size |
|---|---|---|
| Automatic tagging | Once per new or changed document | Small — one short response |
| Conversation summarising | Once per active thread | Small |
| Chat | Only when a person asks something | Depends on the question |
Keeping it low
Leave automatic tagging off unless you want the taxonomy. It is off by default, and it is the only thing that scales with corpus size rather than with usage.
Use a small model. These tasks do not need a frontier model — they are classification and summarisation over material already retrieved.
Remember search is free. Agents doing heavy research cost you nothing on the LensHub side; the cost is in whatever your agent runs on.
The one to watch
A first sync of a large corpus with automatic tagging on is the only operation that generates a large number of model calls in a short window. If you are turning it on for an existing corpus, expect a burst.