Local document chunking for retrieval experiments

RAG Chunking Playground

Split long text by words, characters, tokens, or paragraphs. Tune overlap, inspect chunk stats, and export JSONL, JSON, or Markdown for embeddings and evals.

Local processingWords / chars / tokensJSONL / JSON / Markdown

RAG chunking

Test chunk size, overlap, and export shape before embedding.

Chunk byChunk sizewordsOverlap17% of chunk size

Export as

Total tokens80

Chunks1

Avg/chunk80

Overlap20 words

Source document

64 words · 433 chars

Output

1 chunks · 80 avg tokens · JSONL

Use words for readable review, tokens for model sizing, and paragraphs when structure matters.

JSONL works well for embedding jobs, batch scripts, and eval fixtures.

Chunk 164 words · 431 chars · 80 tokens

AidenGPT is a browser-first utility hub for writing, SEO, text analysis, and developer workflows. RAG systems work best when documents are split into chunks that preserve useful context. Chunk size and overlap affect retrieval precision, answer quality, and embedding cost. Short chunks can improve retrieval precision, but they may lose context. Longer chunks preserve meaning, but they can add noise and increase embedding spend.

Capabilities

What this tool handles

Use RAG Chunking Playground to check the input, tune the settings, and copy a clean output.

Chunk by unit

Choose words, characters, token-style chunks, or paragraphs depending on your retrieval setup.

Tune overlap

Adjust context carryover so chunk boundaries do not drop important meaning.

Export for pipelines

Copy JSONL, JSON, or Markdown for embeddings, evals, notes, or debugging.

Workflow

Run it in three steps

Paste source text, choose a chunking strategy, then copy structured output for your next step.

Paste source text

Drop in documentation, transcripts, Markdown, or any long-form text.

Choose chunk settings

Set unit type, chunk size, and overlap while chunk counts update.

Copy structured chunks

Export JSONL for embeddings, JSON for apps, or Markdown for review.

Good for

Common jobs this tool supports

Testing chunk sizesCompare retrieval boundaries

Preparing embeddingsShape batch-ready chunks

Debugging overlapCheck context carryover

Creating eval fixturesCopy structured test data

FAQ

Answers before you run it

Practical notes about output, privacy, and common settings.

Start with 300-800 words or roughly 500-1,200 tokens, then test retrieval quality against your actual questions.

Use next

Text File MergerCombine source files before processing.LLM Context TrimmerFit long text into model limits.Token Speed VisualizerEstimate generation time before a run.JSON Schema GeneratorShape structured output contracts.