Understand what counts toward context
Context includes more than the latest message. Instructions, available tool descriptions, memory, previous messages, and tool results all contribute to the amount the model can consider in a turn. A large file result or a long conversation can therefore use context even when your current prompt is short.
The effective window depends on the selected model and its declared capacity. IaGenify can use a requested `contextTokens` value when it is valid for that model; asking for more than the model supports produces a context error. The displayed gauge helps show how much of the configured window is currently occupied.
Keep a long task useful as it grows
When context usage gets high, older tool results may be cleared and earlier conversation may be summarized to make room. This can preserve the main task and conclusions, but small details from earlier messages may no longer be available verbatim. Restate identifiers, constraints, or exact text that a later step must use.
You can compact a supported session to summarize earlier exchanges deliberately, or start a new task with a concise handoff containing the goal, decisions, and required files. For work that repeatedly exceeds the model's window, narrow the source material or split the task into smaller stages instead of setting a context limit above the model's capacity.