llnl / llnl/mada

Add support for the Magentic Orchestration Pattern

Open
#13 0 comments 0 reactions 1 assignee Claimed by @Mamorri View on GitHub
enhancement
Dominant language
Python
Stars
2
Forks
2
Avg merge
1d 4h
Merged PRs (30d)
4

Description

The current planner is a centralized agent-as-tool pattern. A single PlanningAgent owns the conversation, decides what to delegate, and invokes specialist agents as tools. Specialists do not coordinate directly with each other; they are exposed to the planner as callable units with bounded inputs. This keeps control flow simple, preserves the current interface contract, and makes startup/participant validation straightforward.

The planned **magentic** pattern is a peer-style orchestration model rather than a single-planner façade. Instead of one planning agent wrapping all specialists as tools, the orchestration layer would allow agents to participate more natively in a coordinated multi-agent exchange. That shifts the system from “planner delegates to tools” to “agents collaborate under an orchestration policy.” In practice, that means different routing semantics, different message/control flow, and likely different expectations around turn-taking, shared context, and coordination strategy. It is more flexible, but it also introduces more moving parts and a larger behavioral surface area than the current planner.

Contributor guide

No contributing guide indexed for this repository

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.