οὐθείς
EN · DE
Overview OS Principles Systems Survey Architecture Data Formats Related Work Agent Prompts Hybrid Memory Stack Quality Threshold

Related Work

This document surveys existing research on applying operating system principles to AI agent architectures.

The Emerging Field

The application of OS concepts to AI agent systems is a nascent but growing research area. Recent work recognizes that agent architectures face challenges operating systems solved decades ago.

As one recent paper observes: "Today's agent architectures resemble the pre-OS era of computing—a chaos of duplicated solutions lacking fundamental abstractions for resource management, isolation, and coordination."

Key Publications

AIOS: LLM Agent Operating System

Source: arXiv 2403.16971, COLM 2025

The most directly relevant work. AIOS proposes an OS kernel for LLM-based agents with:

Key insight: LLMs are treated as cores, analogous to CPU cores, with a unified interface for diverse LLM endpoints.

Difference from outheis: AIOS focuses on multi-tenant agent serving with performance optimization. outheis focuses on personal assistant with privacy guarantees and plaintext data architecture.

Agent Operating Systems (Agent-OS)

Source: Preprints.org, 2025

Proposes a layered architecture:

  1. Kernel plane
  2. Resource & Service plane
  3. Agent Runtime plane
  4. Orchestration & Workflow plane
  5. User & Application plane

Emphasizes real-time guarantees and security primitives for autonomous systems.

Difference from outheis: Agent-OS targets enterprise/SmartTech scenarios with formal verification. outheis targets personal use with simplicity and transparency.

Multi-Agent Memory from a Computer Architecture Perspective

Source: arXiv 2603.10062, 2026

Frames multi-agent memory as a computer architecture problem:

Key insight: "Agent performance is an end-to-end data movement problem."

Relevance to outheis: Validates our index-based access strategy and the distinction between hot (messages.jsonl) and cold (archive/) storage.

Integrating AI into Operating Systems: A Survey

Source: arXiv 2407.14567, 2025

Comprehensive survey covering two directions:

  1. AI for OS: ML/LLM techniques to enhance OS (scheduling, memory, security)
  2. OS for AI: OS architecture innovations to support AI workloads

Identifies three paradigms:

Relevance to outheis: Confirms the validity of applying OS principles to agent design; provides vocabulary and framing.

Modeling an Operating System Based on Agents

Source: Springer HAIS 2012

Early work proposing OS modeling with multi-agent paradigms, considering interaction-based computing and cloud computing.

Relevance to outheis: Shows this is not an entirely new idea, but predates the LLM era.

The Orchestration of Multi-Agent Systems

Source: arXiv 2601.13671, 2026

Technical blueprint for enterprise multi-agent systems:

Difference from outheis: Focuses on enterprise orchestration with complex protocols. outheis uses simple message passing with append-only queue.

Conceptual Parallels

OS Concept AIOS Agent-OS outheis
Kernel LLM kernel with modules Layered planes Dispatcher (no LLM)
Scheduling FIFO, Round Robin Real-time guarantees Priority + keywords
Memory K-LRU eviction Three-layer hierarchy Index + lazy load
IPC System calls Protocols (MCP, A2A) Message queue (JSONL)
Access Control Privilege groups Security primitives Capabilities (pledge/unveil)
Context Switch Logits-based snapshot Not specified Conversation archival

What Distinguishes outheis

1. Privacy-First Architecture

Most related work assumes multi-tenant cloud deployment. outheis targets two modes: Personal Assistant (single-user, local-first) and Domain Expert Assistant (specialized knowledge service). Both prioritize privacy:

2. Plaintext Data Philosophy

Related work typically uses databases or specialized storage. outheis uses:

This draws from Unix philosophy rather than database tradition.

3. Simplicity Over Performance

AIOS optimizes for throughput (2.1x faster execution). outheis optimizes for:

4. Personal Agency

The Agenda agent concept—enabling user agency through intelligent filtering—is not present in enterprise-focused related work.

Theoretical Foundations

The related work draws primarily from:

outheis additionally draws from:

Open Research Questions

The literature identifies several open problems relevant to outheis:

  1. Memory consistency in multi-agent systems
  2. Cache sharing protocols across agents
  3. Context management for long-running conversations
  4. Tag harmonization via LLM (addressed in our theoretical work)

References

  1. Mei, K., Li, Z., et al. (2024). AIOS: LLM Agent Operating System. arXiv:2403.16971. COLM 2025.

  2. Agent Operating Systems (Agent-OS): A Foundational Specification. Preprints.org, 2025.

  3. Multi-Agent Memory from a Computer Architecture Perspective. arXiv:2603.10062, 2026.

  4. Integrating Artificial Intelligence into Operating Systems: A Survey. arXiv:2407.14567, 2025.

  5. Cámara, J.P., et al. (2012). Modeling an Operating System Based on Agents. HAIS 2012, Springer.

  6. The Orchestration of Multi-Agent Systems: Architectures, Protocols, and Enterprise Adoption. arXiv:2601.13671, 2026.

  7. Ge, Y., et al. (2023). LLM as OS, Agents as Apps: Envisioning AIOS, Agents and the AIOS-Agent Ecosystem.

Conclusion

The application of OS principles to AI agent architecture is an active research area. Existing work focuses primarily on performance optimization for enterprise multi-agent systems. outheis contributes a complementary perspective: privacy-first personal assistant architecture with plaintext data philosophy and emphasis on user agency.

The theoretical foundation linking prospective information architecture to agent design (see: Temporalization of Order) appears to be novel and not addressed in current literature.

Open source under AGPL-3.0 · Code on GitHub