Chunk — RAG Chunking & Retrieval Planner

Chunk — RAG Chunking & Retrieval Planner — a premium, 100% offline digital tool by Aceplore. Runs entirely in your browser: no accounts, no subscriptions, your data never leaves your device.

A private, offline planning bench for retrieval-augmented generation — how many chunks a corpus becomes, what the overlap duplicates, what one embedding pass costs at the rate you enter, and whether top-k chunks plus your prompt still fit once the output reserve is taken out — 32 tools in one, and an overlap at or above the chunk size is refused rather than costed.

32 built-in tools in one app. Corpora and documents sized either from a figure you know or estimated from document count and average character length, with a coverage view comparing the documents you have listed against the corpus total you entered; chunking strategies made of a chunk size, an overlap and a split rule, a chunk count worked line by line as ceil((N − V) ÷ (C − V)) with a table of where the cuts land, overlap cost as (chunks − 1) × V and what share of the corpus that repeats, storage overhead reported as a percentage AND in tokens under either counting convention, a boundary model for token, sentence and paragraph cuts that is labelled a model, a small-chunk view against a floor you set, and the remainder chunk per document; a compare table putting every strategy over the same corpus and a chunk-size sweep across a range you choose; indexes carrying your own embedding rate and dimension count, embedding cost, re-index cost when the strategy changes, and a vector-storage estimate built from editable multipliers; retrieval configurations holding top-k, prompt, output reserve and window, with the retrieval budget, window fit, a top-k sweep, an output-reserve view and a truncation order; plus CSV exports, a printable plan report and a method page that works every formula on your current numbers.

Chunk count. 50,000 tokens at C = 800 and V = 100 gives a stride of 700 and 72 chunks. 71 would leave the tail of the corpus unindexed, so the count rounds up — and the table shows where every cut lands.

Embedding cost. stored ÷ 1,000,000 × your price per 1M. 57,600 tokens at 0.13 per 1M comes to 0.007488; at 0.26 it doubles. The rate is read from the index you set up — never assumed, and no price list ships with the app.

Retrieval budget. top-k × chunk size + prompt, checked against the window MINUS the output reserve — the reserve has to come off first. You get a verdict and, when it misses, the shortfall in tokens.

What's included:

• Chunk.html
• Quick-Start-Guide.html
• How-To-Use-Guide.html
• README.txt
• LICENSE.txt

Includes a personalized welcome wizard, an interactive guided tour, a ⌘K command palette, light/dark themes with 8 accents, and one-click JSON backup & restore.

Chunk is a planning, costing and record-keeping tool for people building with AI. Every figure it reports is computed by a stated formula from the rates, sizes, counts and thresholds YOU enter. It ships no model prices, no context windows and no tokeniser: those differ by provider and model and change without notice, so you enter them and the app shows which ones it used. Any token figure derived from the length of a text is an ESTIMATE, not a count — real tokenisers are model-specific and cannot run offline — so treat it as a planning figure and confirm actual usage against your provider's own reporting and billing before you commit to a budget. Nothing here is a guarantee of a cost, a saving, a quality level or a result, and none of it is business, financial, legal or engineering advice. You are solely responsible for your own decisions.