jmchilton [11:35 AM]
Okay - I have a new branch where pages/reports and “Galaxy Notebooks” share a common toolbar and set of database models (in addition to previous common syntax, editor, and rendering pipeline) - https://github.com/jmchilton/galaxy/tree/history_pages. This means pages get the revision mode, chat interface, etc… implemented in the context of “Galaxy Notebooks”. My thinking is they are all just kinds of pages on the backend but the frontend decorates the UI as having “Galaxy Notebooks” (history attached pages on the backend - just page documents with a history_id) and “Reports” (previously pages - just… pages). We already have backend concept vs frontend branding differences for file sources and object stores. But I did have Claude consolidate all the frontend naming and tooltips for both into a single file (https://github.com/jmchilton/galaxy/commit/623ff7e22fb293b6c1175a2e7d91876cf02dcc33#diff-e938dc4d84f94d923890eef95[…]4a0c0b436ddcd12dd086b676033c464) so we can just bike shed this very rapidly. We can revert the concept of a “Report” back to Pages or to “Published Notebooks” or “Persisted Notebooks” or whatever from one file.
nekrut [11:46 AM]
@jmchilton -> I dying to test this. Can we set up a dev insatnce or, I suppose, I can run this locally?
Marius van den Beek [11:48 AM]
We can run this from test
nekrut [11:50 AM]
let’s do this!
nekrut [12:05 PM]
@jmchilton -> do you have move examples / descrioptions/ anything for the notebooks? Just to get a precise idea?
[12:05 PM]
in addition to this: https://github.com/galaxyproject/galaxy/pull/21943
nekrut [12:18 PM]
also this: @HISTORY_MARKDOWN_ARCHITECTURE.md
jmchilton [12:26 PM]
I don’t have examples - but your current love of that external Claude talking to Galaxy is a sort of a driving user case. If your Claude was using the MCP to write Markdown to a history-attached notebook instead local Markdown files - you could have embedded visualizations, embedded dataset views, tabular and image views, interactive job parameter tables etc… Presumably we can use something a lot like the chat agent in this PR to create a Claude skill to teach your Claude the Galaxy syntax and then we do a visualization agent in Galaxy and extract it as a Claude skill and let you do that also (visualization rough draft plan here https://gist.github.com/jmchilton/59c11f4a4422c3b4bffd1816b5f8f4fc). I’m imagining your flow would be something like “Do an analysis like paper X on datasets a,b, and c - as you debug things keep your final results and how they were generated updated and documented in a history notebook.” Then Claude would go fuck a bunch of stuff up but give you a history with a lot of failed experiments but final results summarized in a history notebook. You would be able to look at that interactively - with rich visualizations and such - and then refine what you ask Claude for. “Redo the analysis but adjust bowtie to use parameter Y and then filter the table in step 80 on criteria XYZ, update the notebook to reflect the new analysis.” Claude would run off and do that and you’d get an updated summary. So you’re happy because the viz and such is all embedded - it is a richer view of your analysis to have everything tied together in the UI and I’m happy because I can extract a workflow from your notebook in a way richer way than picking individual jobs and datasets in the current workflow extraction mode and that extracted workflow will be complete with a report derived from your notebook. We’re keeping people who love driving an analysis with Claude in Galaxy by giving them richer visualizations of these agentic analyses, we’re keeping visualizations and summaries tied to the data and closer to the data, and we’re making everything more easily and more richly reproducible than ever before (as Claude nicely summarized one time - we’re making not just the analysis reproducible but the communication of it as well). I’m sorry I have nothing concrete - I never do but the ideas typically work out. We’ll get something broken merged - Marius will use it and make it awesome for some random project - and then we will have the example and it will be great.