#Onboarding improvements 🧵
strong stratus
We need to improve the onboarding experience... Starting a thread to coordinate the various tasks in flight.
@vagrant mural @slim niche @vale totem @bitter lodge
bitter lodge
I think one of the most important pieces is an AI quickstart.
strong stratus
I agree but we should do it in small end-to-end increments
can't afford a 4-week project based on today's UI, that will probably be out of date by the time it ships
slim niche
yeah maybe we can get a short walkthrough of toy-programmer that can act as a quickstart for now
bitter lodge
Yes that's one of the problems I'm having as well, it's moving so fast that it's hard to figure out what to cover
bitter lodge
yeah maybe we can get a short walkthrough of toy-programmer that can act as a qu...
something like
- intro
- run it, see the output, see the trace
- explain workspace module
- explain agent module
- link to other examples
vale totem
I also think its worth a quick touch on "what is an agent" for those interested - even if its a link to another page/site
strong stratus
I think right now the most urgent:
- Fix papercuts in current content
- Build example modules and record videos from them
- Test the new llm-multiobject branch, which merges dagger shell & llm capabilities, and adds support for composiing multiple modules in the llm. Unblocks new modes of usage which we need to experience first hand
slim niche
so on the docs side the idea is (for now) reduce friction and point to examples / videos instead of building up guides?
can we list out the current papercuts in this thread?
bitter lodge
My worry here is that if a user sees a video, thinks "that's cool, I want to try", comes to our docs... He doesn't have any step by step guide to learn basic concepts so he can build one of his own
strong stratus
My worry here is that if a user sees a video, thinks "that's cool, I want to try...
That sounds like a papercut 🙂 Just have to focus on filling that gap incrementally IMO
Maybe instead of the one and only canonical quickstart, focus on example explainers?
Pick an example module that already exists; and write a guide that explains it?
less risk of writing stale content that way
bitter lodge
Another one we should fix is the features section, it's currently heavily focused on the CI use case. The feature section doesn't match the new list of features we have in the readme and on the main docs page.
strong stratus
yes agreed
bitter lodge
Pick an example module that already exists; and write a guide that explains it?
Yes I think this is the same as the short walkthrough of toy programmer Kyle suggested above
bitter lodge
A related question then is: should this be the (temporary) AI quickstart in the docs, or should this be added to the readme or similar doc for each example repo?
FWIW I'm in favour of putting it in the docs so it's more visible
vale totem
Agree, walking through an existing module/agent would definitely be a really good first start
strong stratus
I'm fine with either. Just be mindful of keeping everything up-to-date
bitter lodge
One other papercut until the llm branch is merged: the default installation path in the docs is different. So depending on your system path setup, you might install the llm release but dagger version still points to the older one. My co-presenter ran into this today.
bitter lodge
I think we should. And have the same path, adding a note in the llm install that this overwrites the existing dagger installation
strong stratus
Works for me (note you'll have to remove the homebrew notes I think)
strong stratus
One paper cut in the docs: list of model names that are known to work (the actual strings to give the llm function)
I want to try claude but don't even remember what I'm supposed to enter
slim niche
I'm going to take a quick pass at adding something after dinner to supplement the examples that can serve until we have a basic tutorial. I'm thinking a 'common design patterns' (or something) section that briefly covers
- the agentic loop (the part that happens inside dag.llm().withprompt())
- objects -> tools
- the workspace construct
- getting stuff out of the llm + laziness
- prompting tips / faq
It'll be something quick and small that we can start actually iterating on. I feel like we're stuck on step 0!
strong stratus
Yeah I love that approach - thanks!
slim niche
https://github.com/dagger/dagger/pull/9869
I also updated llm.8 -> llm.9 and removed the unnecessary OPENAI_API_KEY for ollama configuration
Page is getting a bit long, I think next time we add a bunch we should break it up like
Overview
|- Installation
|- Configuration
|- Examples / Tutorial / FAQ
I think the "Integrate Dagger into your application" doesn't fit now that we've added a bit. I'm not sure what the original intent was with that section
slim niche
It's on netlify, let me see if I have permissions
Looks like I don't 😞 I'll sort that out monday. Flow is:
- https://app.netlify.com/sites/docs-dagger-io/overview
- click top item under "Production Deploys"
- Publish Deploy ( + confirm)
vagrant mural
slim niche
slender ginkgo
Looks like I don't 😞 I'll sort that out monday. Flow is:
- https://app.netlify....
I have access, so let me know if you ever need me to click the button
slim niche
I have access, so let me know if you ever need me to click the button
Thanks! I have access now as well 🚀
strong stratus
@slim niche should we also close this thread?
slim niche
yep!