@Slava Tykhonov thanks for the interesting presentation today on https://ai.codata.org ! I posted the recording and your slides on DataverseTV.
Thanks also for the external tool manifest file for ai.codata.org (Ghostwriter 2.0), which I just added to https://demo.dataverse.org in the usual way:
curl -X POST -H 'Content-type: application/json' http://localhost:8080/api/admin/externalTools --upload-file manifest.json
I'm playing around at https://demo.dataverse.org/dataset.xhtml?persistentId=doi:10.70122/FK2/MBVSRE and launching the tool by clicking "Access Dataset" and then "Ghostwriter 2.0".
![]()
The tool launches fine. It's ready for questions.
![]()
I know my tabular file has info about the number of stars per GitHub repo so I asked about that.
But the tool is telling me it doesn't have the data, just the metadata.
![]()
"Based on the provided text, the document describes the contents and schema of a dataset containing information on Harvard repositories (including stars, language, and repository URLs), but it does not provide the actual data or list the specific repositories and their star counts. Therefore, the provided context does not contain the information needed to determine which repo has the most stars."
Am I using the tool wrong? Can it download data files and tell me about them?
Hi Phil, there is some default limitation of the local model. We need to activate it soon to get cloud Ollama subscription to serve all users.
@Slava Tykhonov ok, but am I using the tool wrong? I guess I was wondering if it could behave like AskTheData, which can operate on data files, tabular files, spreadsheets.
Basically, I'm looking for a story to tell that starts like this:
Can you try now? It's connected.
It works!
![]()
![]()
@Slava Tykhonov should we enable this on Harvard Dataverse? You want traffic to ai.codata.org?
I can put it on ask.dataverse.org, no worries.
Sounds fine! No rush! Great tool! ![]()
Deployed on Ask Dataverse now: https://ask.dataverse.org/vault/doc/w3xU4QkBNjnrQLz4yCiYRQ
And look, it can run Agentic AI, download files and respond to questions!
![]()
Wow! Ok! Thanks! ![]()
We should post an update to #community > ask.dataverse.org. To explain how it's changed.
![]()
Works with Claude and other AI models
Can you try it as Dataverse external tool?
Sure! What should we call it? :thinking: I'm not so sure about Ghostwriter...
![]()
@Ceilyn Boyd and I were joking that it should have an animal name. Ferret or Honeybadger or something. That way, it could do anything! Limitless features!
Ask Dataverse?
Hmm, we do already have AskDataverse at https://github.com/IQSS/askdataverse
But it's a bit deprecated, perhaps. :sweat_smile:
Ok, let's call it Semantic Croissant then?
I think I like Ask Dataverse better.
Or Honeybadger. :smile:
Ok, added to demo.dataverse.org as "Ask Dataverse".
You can try it at https://demo.dataverse.org/dataset.xhtml?persistentId=doi:10.70122/FK2/MBVSRE for example.
![]()
@Slava Tykhonov I changed it to "Ask Dataverse (AI)" (suggestion from @Leo Andreev):
![]()
How does that grab you?
![]()
![]()
Look, it's really actionable!
I am your ambassador, Slava! I make it so people can click something and experience your great work! :smile:
Well, we know a little bit how to do AI in Europe as well! ;)
I'll say!
https://github.com/codata/semantic-croissant/blob/fb56de50ab1d48ea2dbc81f22f4e22af231af520/manifest.json has "Ghostwriter 2.0" in it and points to ai.codata.org.
Do you want to provide a second manifest file called "Ask Dataverse (AI)" that points to ask.dataverse.org?
Yes
Should I put the fork with source code on gdcc? To get community contributions.
Sure! Sorry, where's the fork?
Last updated: Nov 02 2026 at 07:01 UTC