Website guide / Source linked

Ask Maviter

Find prompts, compare tools and navigate Maviter. AI answers stay inside Maviter's indexed evidence.

Search is freeSign in for five daily AI credits
Never paste passwords, API keys or confidential project material.
Maviter.
Ask Maviter
Menu
Improve a prompt

RAG Chunk Planner

Estimate document chunks, repeated overlap and the retrieved passages that fit alongside your instructions and answer.

Use the tool ↓
  1. Split the source
  2. Retrieve useful passages
  3. Reserve answer space
Free · Works in your browserLocal by default. Only an explicit save sends numbers/choices to your private account; refreshing clears unsaved calculations.
1 / Your inputs

Set up your calculation

Example values are filled in where useful. Replace them with your own.

Use the exact tokenizer count when available.

Assumes equally sized documents. For mixed lengths, plan groups separately.

Suggests an initial chunk size for evaluation; it does not split or embed your files.

Choose 256–1,024 as a starting experiment, then evaluate retrieval.

Must be smaller than the chunk. Overlap increases embedding volume.

Requested maximum retrieval count.

Enter the documented limit for the target model.

Includes tools, system text and conversation history.

A further 10% of the context is reserved for counting variance.

Calculation inputs stay in this tab unless you explicitly save an eligible snapshot. No account, credit or model call is needed to calculate. Pasted-text tools always stay local.

2 / Your result

Your answer will appear here

Set the inputs, then select “Plan retrieval capacity”. You will see the result, its assumptions and the next useful step.

  1. Split the source
  2. Retrieve useful passages
  3. Reserve answer space
Understand the answer

Method and worked example

Chunks per document = 1 + ceiling(max(0, source tokens − chunk size) ÷ (chunk size − overlap)). Repeated overlap is charged only between chunks. Retrieval capacity = floor((90% of context − instructions − output) ÷ chunk size).

What this result can tell you

No embeddings, tokenizer or relevance evaluation runs. Preserve tables, code blocks and source identifiers; check retrieval with known-answer questions.

Check the complete context budget →

Method version 3.16.0

Common questions

Is one chunk size best for every document?

No. Short FAQs, narrative documents and code need different boundaries. Treat the size as an experiment and measure whether answers retrieve the right passage.

Does overlap reduce the answer budget?

Retrieved chunks can contain repeated text. This planner budgets their full size so overlap is not treated as free context.

References and next checks

Documentation checked 2026-09-05. Provider capabilities can change; use the linked documentation for the exact model or plan.