Context Window Planner
Drop the files you're about to paste into an AI session. See how much of the context window they'll eat — per file — and whether you'll have room left for the actual conversation.
Same tiktoken engine as the Token Counter. Same caveat: these are estimates based on OpenAI's cl100k tokenizer. Different providers use different tokenizers — counts can differ by 10-20%, especially with code. You'll get the right ballpark, not the exact number.
Tap to browse or drop files here
.txt .md .json .yaml .py .ts .js .html .css .csv — up to 20 files, 5MB each
add files to see how they fit in your model's context window
The context window isn't a hard wall
It's more like a room. The more stuff you put in it, the harder it is for the AI to find what it needs. A model with 200K tokens of capacity doesn't perform the same at 90% full as it does at 30% full.
This is called context degradation — the quality of the AI's attention drops as the window fills up. It's not that the AI can't see your first file anymore. It's that it pays less attention to it, the same way you'd skim a 50-page document but read a 5-page one carefully.
Why "reserve for response"?
The context window is shared between your input and the AI's output. If you fill 95% of the window with files, the AI has 5% left to answer you. That's why this tool lets you set a reserve — the breathing room the AI needs to actually think and respond.
The default reserve is 4,000 tokens — about 3,000 words. For complex code generation, you might want 8,000-16,000. For a quick question about a file, 2,000 is fine.
The ~70% Guideline
A practical observation from heavy use: performance tends to stay sharp when you keep your input under about 70% of the context window. Above that, models start losing track of details more often — not always, but enough to notice.
This isn't a published finding with a paper behind it — it's a working guideline. Some tasks handle high context just fine. But if you're getting worse output than expected, your context budget is the first thing to check.
Same deal as always — your files never leave your browser. The tokenizer runs locally. I can't see what you upload, and I don't want to.