Question your archives
Ask a question in plain language. Joha answers only from your documents, every sentence links to its original page or photograph, and the gaps are flagged.
Your archives, questioned. Every answer sourced.
Letters, inventories, scans, photographs: Joha turns your institution's archives into knowledge your teams can question, explore and publish from. Every sentence links back to the page or photograph it comes from.
Built for the rigour of a collection
Nothing without a source
A sentence the AI cannot tie to a passage is removed, not shown. What the archive does not say is stated plainly.
Your team decides
Tags, duplicates, authority links: the AI proposes, a person approves. Every significant change takes a backup first.
Your data stays yours
A project is a plain folder on your computer or shared drive. Only passages leave, one at a time, to the AI provider you choose, or to none with a local model.
From the archive to the exhibition
Drop in your documents, run the processing, then ask your questions. Joha does the sifting, your teams keep control over what enters the collection.
Find exhibition angles
Give a theme and Joha proposes story threads your collection can carry, each backed by at least two cited passages. Keep the best one as a dossier in one click.
See the connections in the graph
People, places, events and objects are extracted from your documents and linked together. Every link points to its source and can be tied to Wikidata, IdRef, the BnF or VIAF.
Put everything in order
The timeline puts documents and photographs back in time, down to the day when the sources allow it. Every date the AI proposes is flagged and can be checked.
The AI proposes, your team approves
One Review inbox gathers proposed tags, duplicates and uncertain links, each with its confidence. Nothing enters the collection without approval.
From dossiers to gallery labels
Gather evidence in a dossier, then draft a gallery label, room panel or collection record, every sentence sourced. Export to CSV, JSON, Dublin Core or Markdown.
One button to process everything
PDFs, scans, text and photographs: Joha reads the scans, describes the images, cuts them into passages, finds people and places, dates them and proposes tags. One click on Run all is enough.
For museum and heritage teams
Co-create exhibitions
Bring unexpected angles out of your archives and build the narrative with the evidence at hand.
Bring a collection to light
Make a little-explored collection readable and searchable by the whole team, without months of sifting.
Write labels and records
Start from a dossier of evidence to write labels, panels and records, every claim sourced.
Frequently asked questions
Where are our archives stored?
In a folder on your computer or shared drive. Nothing is hosted by us. Only passages are sent, one at a time, to the AI provider you choose (Google, Anthropic, OpenAI, Mistral), or to none if you use a local model.
Can the AI make things up?
Joha removes any sentence it cannot tie to a passage of your documents, and states explicitly what the archives do not say. Proposals (tags, dates, links) wait for a person's approval.
Which documents can Joha read?
PDFs (text or scanned), text files and images (photographs, scans). Scans go through text recognition and photographs are described.
Do we need technical skills?
No. Drop in your documents, click Run all, then ask your questions. Fine tuning stays tucked away in an expert mode.
Can several people work together?
Yes. Put the project on a shared drive: comments, questions and links carry their author's name, and Joha tells you when a colleague has the same project open.
How do we get our data out?
At any time, as CSV, JSON, Dublin Core (for collection databases, Omeka or an OAI harvester) or cited Markdown dossiers.
Let's see what your archives have to say
A 30-minute demo on a collection close to yours. Tell us in a few words what you hold and what you would like to do with it.