What question did this study set out to answer?

The aim is to improve orchestration efficiency in multi-agent AI systems through a new planning architecture.

March 12, 20260 citationsOpen Access

LLM-Driven DAG Planning with Topological Execution for Multi-Agent AI Orchestration

Key Points

The aim is to improve orchestration efficiency in multi-agent AI systems through a new planning architecture.
Introduced OctaMind system featuring a novel orchestration architecture.
Replaced iterative ReAct loop with a single LLM call to create a Directed Acyclic Graph (DAG).
Employed Kahn's topological sort for task sequencing.
Reduced LLM calls by 58–85% on multi-step workflows.
Achieved a topological sort time of <0.1 ms.
Estimated planning success rate for single-agent tasks at ~98%.

Abstract

System Name: OctaMind - Personal AI Assistant SystemThis paper presents a novel orchestration architecture for multi-agent AI systems, specifically the OctaMind system. It replaces the traditional iterative ReAct loop with a "plan once, sort, execute deterministically" pattern. By invoking an LLM exactly once to construct a Directed Acyclic Graph (DAG) and using Kahn's topological sort for sequencing, the system reduces orchestration LLM calls by up to 70% on complex tasks. The architecture features a two-level design: a macro-DAG planner for routing tasks across heterogeneous agents and a micro-DAG engine for individual tool calls within sub-agents. Performance Metrics: Reduces LLM calls by 58–85% on multi-step workflows. Topological sort time: <0.1 ms. Planning success rate for single-agent tasks: ~98%.

Ask AI

Helpful

Bookmark

View Full Paper