Beta Features
Beta means real and usable today, still growing. Each section says what's coming next so you know where it's headed — and what to ask for.
Usage (Beta) — Tools menu, ⇧⌘U
- What it shows — calls and tokens in/out across every conversation, by day and by model, plus the app's live memory footprint.
- Why Tokens in is so large — it adds up every call, and an agent re-sends the whole conversation on each call. A 20K-token chat over 300 calls is 6M tokens with nothing wrong. Look at Avg prompt (per call) and the counter in the composer — current prompt against the model's window — for the number that matters. Naseem keeps that small: finished tool work does not stay in the prompt, so a long task does not get heavier with every step. The total grows only with how many calls a task takes.
- Other agent tools consume the same way — they re-send the conversation on every turn too; most simply don't show the running total. From cache tells you how much of each prompt the provider served cheaply. High is good.
- Accuracy — token counts come from the provider wherever it reports them. Dollar estimates are switched off for now: subscriptions and cached tokens made them unreliable, and a wrong price is worse than none. They come back once they are right. Something look off? Tell us on the Naseem support page.