RAG chunk budget planner

Plan a retrieval-augmented generation chunk budget around a selected model’s context window, system prompt, output reserve, and retrieved chunk count.

How to use this tool

  1. Choose the generation model that will receive retrieved context.
  2. Enter system-prompt and output-token reservations.
  3. Choose how many retrieved chunks to include and use the resulting per-chunk budget.

Worked example: Example: with a 200,000-token context window, a 5,000-token system prompt, a 5,000-token output reserve, and 10 chunks, each chunk can be about 19,000 tokens.

Retrieval budget397,000 tokens
Maximum per chunk49,625 tokens
Total document capacity397,000 tokens

Bring your keys. Keep your savings.