Paste the pieces
A real request to a model is rarely one block of text. It's a system prompt plus documents you looked up plus the chat so far plus the new message. Paste each one into its own box.
Pick a model
Each model has a context window — the maximum number of tokens it can read in one go — and a price per token. The bar shows how full the window is and what the call costs.
Shrink the window
Drag the slider to a smaller window, or a cheaper model, and see which piece gets cut first. That's the question that silently breaks long chats and RAG pipelines.
New here? Hit Load example and drag the window slider down to 8K — watch the chat history vanish.
What's a token, and why does the window matter? Show me an example
A token is the unit a model reads text in — roughly three-quarters of a word in English, so "the meeting ran long" is about five tokens. Every model has a limit on how many it can take in a single request: the context window. 8,000 tokens is a few pages; 200,000 is a long novel.
Say you've built a support bot. Each request is: a 600-token system prompt telling it how to behave, 3,000 tokens of help-centre articles you looked up for this question, the last ten turns of the conversation (2,500 tokens), and the customer's new message (80 tokens). That's about 6,200 tokens in, every single call.
Two things follow. Cost: at a few dollars per million input tokens that's a fraction of a cent per call — until you're doing 50,000 calls a day, at which point it's real money, and the 3,000 tokens of articles are half of it. Truncation: if you move that bot to a small local model with an 8K window, there's room — but by turn twenty of a long chat, there isn't. Something has to go, and if you didn't decide what, the framework decides for you. Usually it throws away the oldest history, which is why the bot "forgets" the customer's order number halfway through.
This page makes both of those visible before they happen: paste the pieces, pick a model, drag the window down and see what's cut.
01Pieces of the request
02Model & budget
Prices are editable placeholders. Vendors change them often — put in the numbers from your provider's pricing page. "Reserve for reply" is the room you keep free for the model's answer (its max_tokens); it comes out of the same window.