Distill
Why it exists

A summary you can't check
is just another opinion.

AI summaries read well, and that is the problem. A confident summary with a wrong number looks exactly like a right one, and the only way to know is to read the whole text again, which is what you wanted to avoid.

Distill numbers every sentence of your text before Claude sees it, and asks for each point to name the sentences it came from. Then it checks every figure in the summary against those sentences, with code, not with another AI.

So you read the summary, and you can still see the source behind every line.
What it does

Short, and traceable.

  • Linked to the sourceEvery point shows the sentences it came from, drawn as lines across to your text.
  • Figures checkedEvery number in a summary is checked against its sentences by code, not by the model.
  • What was left outThe coverage strip shows which sentences made it into the summary and which did not.
  • Saved and searchableEvery summary is kept with its source and checks, found again by title, text or a figure.
  • Your text stays yoursDistill gives no one, its owner included, a way to view what you paste or get back. Export it, or delete one summary or your whole account, any time.
  • Fair limitsHourly and daily limits, and spend caps that pause calmly instead of failing.
How it is built

Built like a real product.

  • AppNext.js 16, React 19, TypeScript, Tailwind CSS 4
  • DataPostgreSQL 16 with row-level security: every query sees only its own person's rows
  • AccountsIts own sign-in: Argon2id passwords, sessions stored only as a hash
  • AIClaude Haiku 5.5 through Anthropic's API, streamed point by point
  • TestsDatabase, unit and browser suites on every push, run against the real container
  • Owner's consoleUsers, activity, spend, roles and brakes, every action audited; a public demo shows it
Who made it

Jerome Badour

A freelance full-stack developer: web apps, APIs, servers and AI features, built to the same standard as this one. Distill is a portfolio project, run for real on its own server. The code is on GitHub, and the owner's console has a public demo.

Work with me on Upwork