Your context map
Define the window, then select a slot to shape its content.
Loading tokenizer…
0 tokens
Filled = content · pale = unused slot budget · red = slot overrun. Footprint uses the larger of budget or content for each slot.
Edit slot
Edit category
Add a category, then add slots to start allocating context.
Counts sum each slot’s literal text with the selected encoding. Message framing, images and provider-specific overhead are excluded. Text is never truncated to fit a budget.
Empty starter template · illustrative budgets · no prompt content yet
Export / import design
gpt-tokenizer 3.4.0 · local BPE counting after the tokenizer loads