Token meter
A live context-window gauge in the workspace header shows how full the agent's context is right now. Click it for session token stats, a per-model breakdown, and a one-click Clear context.
What it measures#
The meter sits in the header of every workspace, next to the status chip. It's reconstructed from the token usage NeuralCode already records on each run: input + cache-read + cache-creation tokenson the most recent turn is how full the model's context window is. The label reads used / window — e.g. 156k/200k — and the bar fills to that percentage.
Fill levels#
The bar changes color as the window fills, so you can see at a glance when it's time to reset:
| Level | Range | Meaning |
|---|---|---|
| Green | Below 75% | Plenty of headroom. |
| Amber | 75–89% | Context is filling up. |
| Red | 90% and up | Nearly full — clear it to reset. |
Session stats#
Click the meter to open a popover scoped to the current context session (everything since the last /clear):
| Stat | Meaning |
|---|---|
| input + cache | Total input, cache-read, and cache-creation tokens this session. |
| output | Total output tokens this session. |
| cost | Dollar cost across the session. |
| runs | Number of agent runs since the last clear. |
Below the stats, a per-model list ranks each model used this session by token count, with its cost and run count.
Clear context#
The Clear context (/clear) button at the bottom of the popover resets the conversation — it runs the same /clear you can type in the composer. The agent starts the next turn with a fresh context window, and the meter resets to empty. Use it between unrelated tasks so stale context doesn't crowd out the work that matters (and inflate cost).