#Launching github.com/dagger/agents
1 messages · Page 1 of 1 (latest)
@gaunt arrow @serene glade @lime phoenix @vestal cliff @eager girder
Here's the plan:
TLDR: move shykes/melvin to dagger/agents, make it the one-stop-shop for everyone interested in building Agents on Dagger.
Tasks:
-
Edit melvin README to fit the new repo @fervent bluff
-
Transfer
shykes/melvintodagger/agents -
Merge @lime phoenix 's ollama section of the README
-
@gaunt arrow @serene glade @woeful trail contribute their example modules to the repo (exact path tbd)
-
Call to action for everyone interested becomes: "visit
github.com/dagger/agents"
Ideally I'd like to avoid the case of dumping all of my demos to dagger/agents/kpenfound/foo. Any other proposed structures?
completely flat will probably too confusing but i could be wrong
yeah. I think namespacing by username is too much. But we can namespace by "mini-project", and each project has its own README + collection of modules?
figure you could do steps 1 and 6 first
Well I want to do a first pass on the README so that at least it doesn't say "this is melvin". But will timebox aggressively
1 and 6 -> agree
each project definitely needs it's own readme. Namespacing by mini-project / standalone demo makes sense but we should also summarize them in the top level README
--> simplified the task list 🙂
@gaunt arrow would you have only one mini-project? Or more?
I think I'll keep mine under examples/melvin
I have:
- multiagent already in melvin
- go-coder which is like advanced toy-programmer
- github-agent that does the automated webhook flow with go-coder
and I'd rename them to something that's easier to understand in dagger/agents
are any of these multi-module, or each is a single module?
maybe the right approach is to have a top-level module, and sub-modules if needed?
(in the case of melvin, I never got around to it, but it would make sense for toy-programmer to be a top-level, and toy-workspace to be its sub-module. Those two are self-contained, could be their own thing
Then separately, I have workspace + github-notifier + reviewer + small parts of demo (which needs to be broken up anyway). I could move that to melvin, or we could try to merge it and go-coder?
+1 to this. Could we have a readme template so people can easily copy and paste to start their own? Kinda like the docs integration page template that Lev created?
updated, re-ready for review
Getting a little stuck on the first paragraph of the new README... 😅 "Dagger is..."
(melvin readme avoided that trap)
Helpful at all?
Wait…no what is Dagger there really 🤣 sorry
Can't we skip that, and just say " This repo contains examples of how to use Dagger for agent use cases. Check them out to learn how to build your own, and join us on Discord".
This repo will be in the Dagger one, so they will get to our old Dagger readme that won't match anyways.
Use this for now:
Dagger provides the execution layer for AI agents — a programmable workflow engine that lets agents dynamically compose and run complex actions in container-native, reproducible environments. Instead of relying on hardcoded API calls or brittle scripts, agents can use Dagger to execute workflows on demand, scaling seamlessly across local, cloud, and edge environments.
With Dagger, AI agents can:
- Go beyond API calls: Run real-world tasks inside secure, containerized environments.
- Orchestrate tool use dynamically: Let agents compose and execute workflows at runtime.
- Eliminate execution failures: Ensure workflows are reproducible and portable across environments.
OK finally done with a first pass (sorry @topaz hatch had missed the above - but can integrate it)
pretty compatible
merging @lime phoenix 's PR then pushing my change
then transfer
cutting it very short!
merged, pushed
not perfect but good enough for now
transfering
@serene glade FYI
@fervent bluff should I also contribute the deployment process of the agent along with its frontend? (the go API that receives the webhooks)
this involves packing the engine and the API on a container, shipping it to fly and calling the agentic module using graphql from the Go code
I didn't for my github webhook module since the webhook side of it is not showcasing the agent module itself, it's just calling it