Research & Publications

Open-Source AI Research

We publish our findings on multi-agent systems, context isolation, and AI harness engineering. All research includes reproducible data and open-source tooling.

White Paper · May 2026

Context Isolation in Multi-Agent LLM Systems

David Scott NovaFuse Inc., Ottawa, Canada 18 pages

We compare three execution paradigms in the Hermes Agent framework: single-session Terminal User Interface (TUI), vanilla Kanban orchestrator with generic decomposition, and Self-Tuned Kanban (STK) with explicit interface contracts. Using a novel Context Isolation Rubric (CIR) across five dimensions, we evaluate 23 experimental runs on code refactoring, research synthesis, and multi-file bug diagnosis challenges. Results indicate that the optimal paradigm depends on task structure — more agents is not always better.

Key Findings

  • TUI outperforms vanilla Kanban on tightly-coupled multi-file tasks (CIS 8.87 vs 7.09)
  • Self-Tuned Kanban prevents catastrophic integration failures but doesn't eliminate mismatches
  • For research synthesis, all three paradigms achieve comparable quality (CIS 8.40)
  • Cohen's d = 1.24, p = 0.10 — not statistically significant, small sample

Need AI expertise for your organization?

We provide consulting on multi-agent systems, AI architecture, and defence procurement. Every engagement is led by a B.Eng with hands-on research experience.

Get in Touch