Sometimes the LLM gives you three paragraphs when you needed three bullets. This tool does the reverse of the expander: it takes verbose prose and collapses it into a tight, ordered bullet list using your own OpenAI or Anthropic key.
How it works
Choose a provider and model, paste your API key, and drop in the long text. Set the maximum number of bullets and pick a detail level — headlines for a quick TL;DR, balanced for one clean line per point, or detailed to keep specifics. When you run it, the tool sends one direct request from your browser to the provider with a system prompt that tells the model to keep only the most important points, in order of importance, without adding anything new. The bulleted summary comes back ready to copy.
Your key is held only in the browser tab and sent straight to OpenAI or Anthropic with the direct-browser-access header (for Anthropic). It is never stored on a server, and refreshing the tab clears it.
Choosing the right detail level
| Detail level | Output style | Typical use |
|---|---|---|
| Headlines | Terse bullets, max 8 words each | TL;DR, slide decks, tweet threads |
| Balanced | One clear self-contained line per bullet | Meeting prep, executive summaries |
| Detailed | One to two sentences per bullet | Digest of a long technical document |
Combine the detail level with the bullet count to dial in exactly how compressed the result should be. For example, 5 bullets at Balanced gives a clean executive summary; 12 bullets at Detailed gives a structured digest without throwing away important specifics.
Getting a good summary
- Match bullet count to purpose — three to five bullets for an executive summary; ten or more for a detailed digest.
- Watch for lossy compression — fewer bullets means more dropped detail. If a must-keep point vanishes, raise the count or switch to Detailed and re-run.
- Feed it clean source text — stray markup or interleaved formatting can confuse the model; paste the prose on its own for the tightest result.
Tips
- Use it as a round-trip with the Bullet Expander: collapse a draft to bullets, edit the structure, then expand back to prose.
- Cheaper models summarize well — reserve premium models for dense or highly technical input where subtle points matter.
- Input length dominates cost here (output is short); trim irrelevant sections before pasting to save tokens.
When to use this versus a prompt-level instruction
If you control the original LLM call, the simplest approach is to instruct the model directly — “respond in 5 bullet points” in the system prompt. Use this collapser when you already have verbose output you did not generate: a colleague’s notes, a previous AI response from a different session, a document you cannot re-prompt. The collapser also works well for iterative compression — if the first pass at 8 bullets still feels too long, reduce to 5 and re-run without touching the original source text.