Skip to content

Tokens and usage

AI work costs tokens. Pavii.Ai makes the spend visible and gives you tools to keep it down.

Usage tracking

Section titled “Usage tracking”

Your profile page shows usage statistics and an activity heatmap, drawn from a per-day ledger the engine records on your own machine — nothing is sent to a server.

Every turn counts, whatever drove it: a chat you are watching, a conversation in another window, a scheduled task, a message from a connected channel, or remote access through the WebUI.

The figures cover everything the model provider processed for a turn, including the cached context an agent re-sends each time. That is why the totals look large next to a chat assistant’s — an agent re-reads its context on every turn, and providers bill for it.

Context visibility

Section titled “Context visibility”

The composer shows how full the current conversation’s context is. When a long session approaches the limit, you can start a fresh conversation (/new) or let agent commands like OpenCode’s /init re-ground the session instead of carrying stale history.

Trimming cost

Section titled “Trimming cost”
  • Token Optimizer — a bundled skill that makes the assistant answer in a compressed, terse style, cutting the tokens it spends on replies by around 65% while keeping the technical detail. It does not shrink your prompts or the conversation context. Enable it on an assistant and invoke it from the / menu; it also responds to “be brief” or “terse mode”. Six intensity levels, from lite through ultra.
  • Token Budget Advisor — another bundled skill, for deciding the spend before it happens rather than trimming it after. It estimates the size of the answer your question is heading for, offers you a set of depth levels with the cost of each, and answers at the one you choose. Ask for the “short version” or say “token budget” to bring it up.
  • Skills discipline — skills load only when you pick them, and a skill already loaded in the conversation is never re-sent, so its cost is paid once, not per message.
  • Free providers — for experiments, route conversations to a free model option in the picker and keep the paid key for real work.