Skip to main content

Token usage and context window

AI models read and write text in tokens. Nexly tracks token usage so you can understand Insights activity and how full the model context is.

Input and output tokens​

Input tokens are the tokens sent to the model. They include your message, relevant chat history, system instructions, selected app context, and tool results that the assistant needs to answer.

Output tokens are the tokens generated by the model in the assistant response.

Together, input and output tokens describe how much model work a turn used.

Cached tokens​

Some providers report cached input tokens separately. Cached tokens are still part of the prompt context, but they may be billed or optimized differently by the provider.

Nexly can track cached reads and cache creation separately when the provider exposes those numbers.

Reasoning tokens​

Some models report reasoning tokens. These represent internal reasoning work reported by the provider. They may not appear as visible response text, but they can still count toward usage.

Request count and turn count​

A single user question can require more than one model request. For example, the assistant may inspect analytics data, receive tool results, and then ask the model to produce the final answer.

Nexly tracks:

  • Requests: model calls made during the chat.
  • Turns: completed user-assistant exchanges.

Context-window meter​

The context-window meter shows how much of the selected model's context window is currently in use.

It is based on the prompt size of the latest or fullest model call in the chat. The denominator depends on the selected model family because different models support different context windows.

The meter can change color as it fills:

  • Neutral for normal usage.
  • Amber near high usage.
  • Red when the chat is close to the model's limit.

Why context grows​

Context grows as a chat accumulates messages, tool results, reports, charts, and previous assistant answers.

If the meter gets high, start a new chat with a focused question and link back to the dashboard view you want the assistant to inspect.

How usage is tracked​

Nexly keeps token usage per chat, so you can see how much model work each conversation used, broken down by model and, when relevant, by the selected app.

Usage tracking never affects your conversations: the assistant answers normally whether or not usage numbers are available for a given turn.