-
Notifications
You must be signed in to change notification settings - Fork 203
Ideal project structure for multiple root agents with some shared tools / skills #50
-
Are there any best practices or guidelines for configuring tools, skills or other components of the eve framework that go across several root-level agents (not sub-agents)? For example, shared tools or skills?
Does defining these separately and importing them into the relevant agent paths work within the file system framework?
Beta Was this translation helpful? Give feedback.
All reactions
Replies: 3 comments 1 reply
-
Shared tools and skills fit well in a shared folder or package, then imported into each root agent that needs them. That keeps each agent self-contained without duplicating code.
The only thing I’d watch is keeping the shared layer generic enough that it doesn’t assume one specific agent state. I can sketch a folder layout that works cleanly for that setup.
Beta Was this translation helpful? Give feedback.
All reactions
-
Reading the release post, it says at the end that Vercel originally used multiple different repos but now all of their agents (over 100, it says) exist in one monorepo. It also mentions there is a routing agent that understands what all the agents do.
To help with orchestration I'd really like to understand the structure of this monorepo. If an example file tree could be shared or someone can point me to more information, I understand that subagents can be used. I feel like the right architecture here is not an orchestrator agent at the root and then 20 or 30 subagents that are completely different functions. It feels more like you have a subagent, like let's say an SDR agent, that has subagents like research that are related to that, and then you'd have another agent like data analytics who has subagents for Postgres. Really great to understand what Vercel are doing here in terms of best practice you
Beta Was this translation helpful? Give feedback.
All reactions
-
Are there any best practices or guidelines for configuring tools, skills or other components of the eve framework that go across several root-level agents (not sub-agents)? For example, shared tools or skills?
Does defining these separately and importing them into the relevant agent paths work within the file system framework?
++ The documentation doesn’t really explain the monorepo structure in detail. For example, are Vercel's internal tools duplicated across dozens of places? If so, how is deployment even handled? Could someone clarify the recommended approach for managing multiple root agents? Spinning up a separate repository for each agent doesn’t seem scalable, especially when many agents share the same skills or tools.
Or the intended setup is to run a single Router agent like Vercel, with all other agents functioning as sub‐agents?
Beta Was this translation helpful? Give feedback.
All reactions
-
🚀 1 -
👀 1
-
to run a single Router agent like Vercel, with all other agents functioning as sub‐agents
This is what I am trying for setting up my personal GitOps agent, the root agent acts as the single router and multiple declared subagents for different GitHub operations. However, the docs say that root and subagents are isolated, this causes unnecessary code duplication among all agents (root and subagents), in the cast that a tool may be common for multiple agents. If a recommended pattern can be documented, it'll be really helpful.
Also, there's a known issue #92 with importing helper module from lib.
Beta Was this translation helpful? Give feedback.