Show HN: Modeling parallel (and hierarchical) agents with Burr
colab.research.google.comHi all, just showing a new construct in Burr (for more info on Burr see - https://news.ycombinator.com/item?id=39917364) for modeling agent / agentic parallelism. We think a "graph" mental model lends itself well to thinking about parallelism (& hierarchy) -- would love your thoughts.
Video walkthrough of notebook - https://youtu.be/G7lw63IBSmY
Why did we build this feature? The future is only going to get more complex with LLMs & agents, and having the right constructs with integrated observability will help us get there. We naturally saw people doing this and thought we could help simplify it more.
Key Features of Burr's constructs:
* you can map over dynamic values in state (e.g. run an agent in parallel over many URLs)
* you can map over dynamic actions/or sub-graphs (e.g. running different agents in parallel)
* you can do the cartesian product between the two (e.g. run over prompts and actions/agents and do it in a loop to modify them...)
More details here - https://burr.dagworks.io/concepts/parallelism/.
Things not mentioned but useful in this parallel context, Burr comes with pluggable durability/persistence/memory (see https://burr.dagworks.io/concepts/state-persistence/).
We'd love to get any early feedback from the HN community, ideas for where you'd like the framework to go, or UI/UX experiences in this space. We will be in the comments for the next few hours to respond! Or leave an issue or discussion here - https://github.com/dagworks-inc/burr. Thanks!
Forgot to mention, on the roadmap we have pluggable executors, e.g. delegating parallelism to whole agents/subgraphs to systems like "ray". Would love to collaborate with folks if that's interesting.