Describe the feature you'd like to request
Claude has this project feature which is ideal for some of the things I have to do: I give it some docs as background and then work, iteratively, on a document, with it using tools and online research if needed etc. It is a very nice and powerful workflow.
It has some elements of integration - like, if it created a document, you can read it there, select some text, and give feedback on that selection or ask it to do something. That would be cool in Text of course.
Describe the solution you'd like
It is very effective to write papers, press releases, migration guides and such. It goes like this, at least, it's how I used it earlier to write a guide for reviewing community conference papers:
- start a 'project''
- give it a name and describe what you want to do
- you get a place where you can upload documents and start a chat
- you can have multiple chats related to the same project. I actually added an existing chat to it, as it was about a related subject.
- I upload documents with background. Examples and such. For example, I uploaded my feedback on conference talk submissions as a markdown table file - after having sanitized it (removed all personally identifiable data using a locally running LLM). And I uploaded a markdown file with general instructions/conference description like I would give a human.
- I wrote a prompt describing what I want: create a guide that can be used by a LLM and by a human to review the talk submissions, doing a first look. (My full prompt was much longer and more detailed)
- I asked it to make a proposal and ask follow-up questions
It said - we should create a guide, a reference document with examples, and a skill file. Bundle those in a .skill file and it can be used by AI while the separate files can be used by a human.
I said - go ahead. It created a good start. I edited and reviewed, giving feedback and editing the file directly, and I got a great end result.
So, a few properties that this project has:
- it has a name and a goal where you can put specific instructions
- it has a workspace ("context") with documents (user can edit/work on them in parallel)
- it can have any number of associated chats
- it has a "progress" section where it tracks a TODO and ticks things off (can be interacted with)
- it has an "output" section with results it generated - but that could just be the workspace, tbh.
- in the chats in also tracks tools and skills it used and websites it visited. I would prefer that to be project-wide, to be honest. Maybe even giving the ability to choose skills and tools pro-actively.
Another factor probably relevant:
The documents uploaded are often not suitable to just be added to context. They might need OCR, or need to be shortened and/or summarized by a (smaller) model first. Example:
So there is some higher level pre-processing and planning needed before going to work on a project ;-)
Describe alternatives you've considered
The current way of working allows you to bring in context into a conversation with the Assistant, using the file picker. But it can't use tools (or skills). Nor can it create a document which can then be iteratively improved. Also, a project could have multiple chats. I can see it as a longer living thing.
Describe the feature you'd like to request
Claude has this project feature which is ideal for some of the things I have to do: I give it some docs as background and then work, iteratively, on a document, with it using tools and online research if needed etc. It is a very nice and powerful workflow.
It has some elements of integration - like, if it created a document, you can read it there, select some text, and give feedback on that selection or ask it to do something. That would be cool in Text of course.
Describe the solution you'd like
It is very effective to write papers, press releases, migration guides and such. It goes like this, at least, it's how I used it earlier to write a guide for reviewing community conference papers:
It said - we should create a guide, a reference document with examples, and a skill file. Bundle those in a .skill file and it can be used by AI while the separate files can be used by a human.
I said - go ahead. It created a good start. I edited and reviewed, giving feedback and editing the file directly, and I got a great end result.
So, a few properties that this project has:
Another factor probably relevant:
The documents uploaded are often not suitable to just be added to context. They might need OCR, or need to be shortened and/or summarized by a (smaller) model first. Example:
So there is some higher level pre-processing and planning needed before going to work on a project ;-)
Describe alternatives you've considered
The current way of working allows you to bring in context into a conversation with the Assistant, using the file picker. But it can't use tools (or skills). Nor can it create a document which can then be iteratively improved. Also, a project could have multiple chats. I can see it as a longer living thing.