For history buffs & lifelong learners
A chatbot is at its worst exactly where your reading gets interesting.
Humy is for people who read history for its own sake. You question a person, an event or a place, each one built out of real documents and searching them before it answers. Every quotation comes back copied out of the page it came from.
Free tier of 20 messages a month · $14.99 a month if a year is too much to commit to
- 1,200+
- historical figures ready to talk, in text or live voice
- 1,323
- quotes in one test of ours, each copied out of its document
- 50 × 1,000
- sources and pages each, in the figures you build yourself
Every AI answer carries its sources underneath. Almost nobody ever opens one.
Readers followed one of those links on 1% of the visits where Google showed an AI summary. Pew Research Center, 2025 (browsing data: 68,879 searches by 900 US adults)
Three things happenbefore a word reaches you.
- 01
Somebody chooses the documents
A person, not a model, assembles the archive behind each figure, and every document keeps its maker, its date, its kind, and where it is held.
- 02
The pages get read before you ask
Scans that are only pictures are typed up a page at a time, old handwriting included, with the scan left beside the text so you can compare.
- 03
The figure goes looking
It lists its archive, searches it, opens a document, reads the page before. Up to 10 rounds or 30 seconds, then it answers from what it found.
“The water was in the parlour before we were awake, and by noon there was not a dry room in Mill Street.”
The AI never types a quote. It marks the passage, our system copies the words straight out of the document, and anything that is not an exact match is thrown away. A quote set out as a quote cannot have been invented.
Ask a figure something its documents do not cover and the answer comes back short, with no quote under it. There was nothing to copy.
The honest numbers
We publish the numberwe would rather not.
1,250 questions about one figure, answered from 17 documents split into 11,117 searchable pieces. It measures whether an answer stays tied to its sources, not what you come away knowing. Answers took 42.3s against 6.4 from memory, at about three times the cost, because it reads the documents before it answers.
- 97%
- more of our benchmark's rubric checks passed than the same model with no documents attached
- 0
- quotes made up, across 1,323 checked against the documents
- 0
- highlights on the wrong passage
- 0
- quotes from outside the figure's own sources
Humy internal benchmark, August 2026: 58.2% of the rubric checks passed with the archive, 29.6% without. Across the 1,323 quotes in that run, one citation opened nothing, and we report it rather than round it away. It was not a controlled head-to-head: the configuration with the archive also ran different instructions and was allowed longer answers.
The sentences around a quote are still written by the AI. Attaching the documents cut outright invention from 22.6% of answers to 12.2%, which still leaves roughly one answer in eight carrying an invented date or reference in that surrounding text. The quote itself is always real. Read the quote as evidence, and the sentence around it as a well-read guide who can still slip. Read the full benchmark.
More than 1,200 people to talk with, and room for the ones only you have.
- 1,200+
The library, ready to talk
World and US history, science, philosophy, the arts. Ask for someone missing and a good part of the library got built that way.
- 50 × 1,000
Or bring your own
Sources and pages each. A name, a short description and the documents, and the siege or the farmhouse can answer for itself.
- 10 min
Say it out loud
Live voice calls, long enough to argue properly while you are cooking. Images and tables stay in text chat.
- 50+
In the language you think in
Text and voice. The documents a figure searches do not change with the language you ask in.
- 2 or more
Let them disagree
Put two figures on the same question. Where the record splits, you see the split rather than a summary.
$129 a year, or $14.99 a month.
One price covers the whole library and everything you build for yourself. Someone assembles each archive by hand and the pages are typed up months before you ask anything, which is the part you never see and most of what you are paying for.
Free
20 messages a month
No credit card. Enough for the one test worth running before you pay: ask about the corner of history you know best, then look at where the quotes came from.
Plus
$129 per year
or $14.99 per month, billed monthly
- The ready-made library of 1,200+ historical figures, in text or live voice
- Up to 50 of your own, built from documents you upload: a person, an event or a place
- Every quote copied out of its document, with that document one click away
- More than 50 languages, spoken and written
- A pool of 12,480 messages a year, or 1,040 a month if you pay monthly
Humy is paid for by schools, universities and museums, and by 143 people who bought it for themselves. That second group keeps the library growing in odd directions.
What people ask us first.
A place to talk with more than 1,200 figures from history, by text or by voice, where the figures answer out of real documents and show you what they drew on. It began as the Hello History app and is now used by schools, universities, and museums as well as by people who simply want to read and argue with the past. 143 individual subscribers pay for it alongside the institutions.
A general chatbot answers history from memory, which is where invented letters and misattributed quotations come from. A Humy figure searches an archive of real documents before it answers, and when it quotes something the text is compiled from the document, not typed by the model. You can open the source and read the passage in place.
Less, and we publish the number. In the benchmark behind Humy 2.6, giving the model an archive cut outright fabrication from 22.6% of answers to 12.2%. That is a real improvement and it is not zero: roughly one answer in eight still contained an invented date or document reference in its prose. The quotation itself is a different matter. All 1,323 verified quotes in that run were exact slices of their sources. Read the quotes as evidence and the prose around them as a well-read guide who can still slip.
$129 per year, or $14.99 per month. There is also a free tier with 20 messages a month if you want to try it before paying. You can cancel whenever you like and keep access until the end of the period you have paid for.
Yes. Live voice calls run up to 10 minutes and work in 50+ languages, so you can hold an actual spoken conversation. Images, tables, and formatted references appear in text chat, not during a call.
Yes. You can build your own figures and attach up to 50 sources to each one, with up to 1,000 pages per source: scanned books, letters, photographs, audio. Scans without machine-readable text are transcribed page by page, historical handwriting included, so a family archive or a local history collection can become something you talk to.
More than 1,200, spanning world and US history, science, philosophy, and the arts. If the person you want is missing you can build them yourself, or write to hello@humy.ai and ask. Requests from subscribers are how a good part of the library got built.
More than 50, in text and in live voice. A figure stays historically grounded whichever language you use, because the archive it searches does not change with the language of the question.
That is what the source panel is for. Every citation opens the document beside the transcription, so you can follow a claim back to the page it came from, compare how two sources describe the same day, and notice where they disagree. Debates between figures are a good way to see a disagreement laid out rather than flattened into one narrative.
Something we have not answered? Write to us at hello@humy.ai.