Split the big stuff into chunks AI can handle.
Two tools in one. Chunk long text into model-ready pieces with token-aware splitting and configurable overlap. Or decompose a complex prompt into a chain of focused sub-prompts that produce dramatically better results than a one-shot.
Paste your long text
Why chunking changes the answer
Two patterns. Both proven in real-world AI work.
Text Chunker — for inputs that don't fit
Long transcripts, contracts, research papers, log files. Pick your target model, set chunk size and overlap, and the tool walks backwards to find natural break points — paragraph > sentence > word > character — so chunks never split mid-thought. Each chunk arrives copy-paste ready with a header or XML wrapper so the model knows where it is in the sequence.
- 9 model presets (GPT-4o, Claude Opus 1M, Gemini 1.5 Pro 1M, etc.)
- Token estimate per chunk + total token count
- Configurable overlap so context doesn't break across chunks
- Three prefix styles: --- header, XML tags, or bare
Task Decomposer — for asks too big for one shot
When a prompt has 4+ distinct deliverables, models cut corners and produce mediocre results on each. A prompt chain — research → plan → draft → critique → polish — consistently beats a one-shot, even with the same model. The decomposer detects complex prompts and emits a tailored chain (knowledge work or code) with each step ready to run.
- Detects complexity signals: verb count, connectives, sentence count
- Knowledge chain: Research → Plan → Draft → Critique → Polish
- Code chain: Spec → Design → Implement → Test → Refactor
- Each step shipped as a stand-alone prompt with role + format
FAQ
›How accurate is the token count?
It's an estimate. We use max(chars/4, words × 1.3) which tracks BPE-style tokenizers within ±10% for English prose. For absolute accuracy, run the chunks through your provider's tokenizer — but the estimate is usually close enough to plan chunk sizes.
›Does the chunker work for non-English text?
Yes — the chunker is language-agnostic. The token estimate is calibrated for English; languages with denser tokenization (e.g. CJK) may use more tokens per character, so use a smaller chunk size as a safety margin.
›What's the right chunk size?
We pre-fill the recommended size for the model you pick (e.g. ~12k for Claude Sonnet, ~32k for Gemini 1.5 Pro). The right answer depends on your task: smaller chunks = more retrieval precision; larger chunks = more context for the model to reason about.
›Does anything leave my browser?
No. The chunker and decomposer are pure JavaScript running in your tab. Your text never reaches any server.