"Why I Ditched One All-Purpose AI for Nine Specialists — A Multi-Agent Design Story"

멀티 에이전트 온톨로지 실전 구현 — 메시지 허브, 컨텍스트 격리, 메모리 설계 2편

A role-based multi-agent architecture that solves context pollution and security vulnerabilities

핵심 요약

  • A single-agent setup failed within three days due to context pollution and security gaps
  • Tasks were split into four categories — coordination, execution, research, writing — forming a nine-agent system
  • Model assignment strategy, fallback chains, and least-privilege access controls tackle both cost and security

Background

When I first set up my AI agent platform, I assigned everything to a single agent. Email checks, web research, code generation, memory maintenance — one agent handled it all.

It failed in three days. Two critical problems surfaced:

  1. Context Pollution — Massive web-crawl results lingered in memory while the agent drafted email replies, causing the model to confuse contexts and waste tokens.
  2. No Permission Boundaries — File-system access for memory management and external web searches require fundamentally different permissions. Giving one agent blanket tool access was a security time bomb.
1. 메시지 허브 — 에이전트 간 통신 설계

The Design

Classification: Four Categories

A clear taxonomy was needed to divide agents. All tasks map to one of four categories:

Category Characteristics Required Tools
Coordination Request analysis, agent delegation sessions_spawn, exec, read, write
Execution Code generation, CLI, automation read, write, edit, exec
Research Web search, data collection web_search, web_fetch, read
Writing Documentation, email drafts read, write, edit

Adding dedicated roles for memory management and monitoring brought the total to nine agents.

3. 메모리 3계층 구조

Nine Agents at a Glance

Agent ID Role Model Tier
mir Coordinator, user interface Premium
monitor System health checks, pattern detection Cheap/Free
researcher Web research, trend scanning Balanced
communicator Content creation, email/document drafting Premium
orchestrator Coding, CLI automation, skill building Balanced
google Google Workspace API execution (Gmail, Calendar, Drive) Cheap/Free
memory-manager Reflect pipeline orchestration Balanced
memory-runner Reflect execution (spawned by manager) Free (Local)
reviewer Monthly memory review and summarization Cheap

Model Assignment Strategy

Different tasks get different models:

  • mir (Premium) — Needs top-tier models for accurate intent parsing
  • researcher (Balanced) — Requires fast throughput for large-volume text summarization, so speed-optimized models take priority
  • memory-runner (Free) — Simple repetitive file I/O runs entirely on a local model

Fallback Chains and Security

Three principles guard against cloud API failures:

  1. Provider Diversification — Alternate across multiple providers so a single outage never takes down the system
  2. Performance-Based Degradation — Start with the highest-performance model and step down to cheaper alternatives on failure
  3. Local Safety Net — A local model serves as the last line of defense, ensuring minimum functionality even during internet outages

Security follows least-privilege: the monitor agent gets only read and exec permissions; the researcher gets web_search and web_fetch but no file-system access. Agent spawning is controlled through explicit allowlists.

Lessons Learned

Too few agents and you get context pollution plus security holes. Too many — splitting "email read-only" from "email write-only," for instance — and management overhead explodes. The sweet spot is dividing by role-sized units: research, memory management, monitoring, and so on.

Takeaway

A nine-agent system organized around coarse-grained roles keeps contexts clean while delivering both security and cost efficiency. Trading the convenience of a single agent bought system-wide stability and scalability.

댓글

댓글 쓰기

이 블로그의 인기 게시물

Agent Memory Engine (2/10) — Building an AI Agent Memory System with SQLite Alone

← 1/10 SQLite Memory Engine f… 📚 Series Index 3/10 memcore Completed → A memory engine that runs without daemons, without dependencies, anywhere 핵심 요약 This post covers how to design an AI agent memory engine (memcore) on a single-file SQLite backend. - Structure: 25 modules, ~6,464 lines, 22 DB tables - Principles: no daemon, single-file portability, no required external dependencies, host-agnostic - Core techniques: 4-layer hybrid search (topic · vector · FTS5 · LIKE), U-tag dialectic-based confidence evolution, ingestion-layer bias rejection rules What This Post Covers One approach to building AI agent memory without a server or cloud vector DB. Specifically: (1) the limitations of text-file-based memory, (2) vector DB alternatives compared and selection criteria, (3) how to layer search tiers, (4) how to evolve confidence scores over time, and (5) how to block biased memory accumulation at the ingestion stage. Design Principles: Why Single-File SQLite Text-file-based ...
Read more »

"ML Foundations (9/9) — PyTorch vs TensorFlow, and the Road to Local LLMs"

이미지
← 8/9 Deep Learning Architec… 📚 Series Index (series end) The final part. So far we've focused on models. Now we focus on the tools that actually run them . We'll lay out the philosophical difference between PyTorch and TensorFlow, and trace how Hugging Face transformers , llama.cpp , MLX , and Ollama built the bridge to running large language models on your own machine. By the end you should have the full mental model of "download a pretrained LLM and serve it locally." 0. Learning Objectives Explain the eager-vs-graph difference between PyTorch and TensorFlow. Explain, in graph terms, how autograd automates the backward pass. Use Hugging Face transformers ' AutoModel , AutoTokenizer , and pipeline abstractions. Describe llama.cpp's GGUF quantization, INT4 inference, and CPU-first flow. Describe Apple MLX's use of unified memory and how it differs from PyTorch. Run a local LLM with Ollama and call it through an OpenAI-compatible API. ...
Read more »

"RAG Core Study (14/26) — Evaluation Sets with RAGAS & DeepEval"

이미지
Series overview: Series index 📚 Series Index Without an evaluation set, every RAG improvement is a story, not evidence. RAG systems fail in more than one place: retrieval, context selection, answer generation, and grounding. That means "looks better" is not a metric. Part 14 explains how to build a golden dataset , how RAGAS and DeepEval fit into the workflow, and why teams must separate retrieval quality from answer quality if they want tuning decisions to hold up. 0. Prerequisites Part 13 reranking — system quality now depends on multi-stage ranking. Part 1 grounding — a correct-looking answer is not enough. Part 12 Hybrid — multiple retrieval settings need comparison on fixed data. 1. Learning Objectives Build a small but useful golden dataset for RAG. Distinguish retrieval metrics from answer metrics . Use RAGAS and DeepEval in the right roles. Avoid the common traps in judge-model-based evaluation. 2. 핵심 요약 An evaluation set for RAG is ...
Read more »

"ML Foundations (8/9) — Deep Learning Architectures: CNN, RNN, Attention"

이미지
← 7/9 Deep Learning Training 📚 Series Index 9/9 PyTorch vs TensorFlow → What did we keep adding on top of an MLP? Vision went CNN, sequences went RNN/LSTM, and eventually both converged on Attention and the Transformer. Each architecture is easier to remember if you read it as what it gave up to gain something else . This part is the one-line summary of the decisive inflection points: LeNet → AlexNet → ResNet → LSTM → Attention → Transformer. 0. Learning Objectives Explain why CNN's convolution, pooling, and weight sharing match images so well. Trace what got unblocked from LeNet → AlexNet → VGG → ResNet as depth grew. State BPTT for RNNs and the vanishing/exploding gradient problem. Write the LSTM gate equations (forget, input, output) with intuition. Write the attention formula in query/key/value form and the scaled dot-product variant. State the precise reasons Transformers replaced RNNs (parallelism + long-range dependencies). 1. 핵심 요약 CNN : weight sharin...
Read more »

"ML Foundations (7/9) — Deep Learning Training: Optimizers, Regularization, Initialization"

← 6/9 Neural Networks 📚 Series Index 8/9 Deep Learning Architec… → Building an MLP, as we did in Part 6, is not the same as getting it to train well . The right optimizer, regularization, initialization, and learning rate — when these line up, deep networks converge. When they don't, the network refuses to learn at all. This part is the catalog of those crafts, with formulas, paper citations, and code in one place. 0. Learning Objectives Compare and write the update rules for SGD, Momentum, Nesterov, and Adam. Explain how Dropout, BatchNorm, and LayerNorm work and where they belong in a model. Derive the variance formulas for Xavier and He initialization and match them to activations. Implement step, cosine, and warmup learning-rate schedules in PyTorch. Explain why gradient clipping is effectively required for RNNs and Transformers. Diagnose the most common training failures (NaN, plateau, overfitting) and apply first-line fixes. 1. 핵심 요약 SGD : \(w \leftarro...
Read more »

OpenClaw to Hermes Migration (2/13) — What to Preserve, Partially Port, or Discard

← 1/13 Current Structure Inve… 📚 Series Index 3/13 What Moves to Hermes → A code review record classifying operational scripts into three tiers: preserve, partial-preserve, and discard. What This Post Covers An asset classification framework (preserve / partial-preserve / discard) for migrating a legacy agent system to a successor A concrete case applying the Strangler Fig + Subset Migration strategy to a memory pipeline Classification results evaluated across LOC, external dependencies, and replaceability axes A method for slimming down a bloated single file (1,310 LOC) using the migration as a natural entry point Problem Definition: Port Everything or Cut Deep? The OpenClaw inventory comprises multiple agents, multiple scheduler/daemon processes, and multiple operational scripts. When moving to the successor system, Hermes, the choice is not binary. A full port carries accumulated technical debt along with it; a minimal port discards hard-won domain logic. Code-revie...
Read more »

AI Agents I Built (5/7) — Building an Automated Blogger API Publishing System

← 4/7 My Life Organizer Agent 📚 Series Index 6/7 Dual → Designing a Markdown-Based Auto-Publish Pipeline with Blogger API v3 + OAuth 2.0 Key Summary Blogger API v3 + OAuth 2.0 enables a pipeline that converts Markdown files into publishable HTML posts blogger_publish.py handles the full flow — frontmatter parsing → body cleanup → HTML conversion → CSS injection → API post — in a single script For bulk publishing, quota management (delay, retry, TOC suppression) is the critical design concern Platform Selection Criteria API access is mandatory for AI agent-driven content publishing. A comparison of platforms: Platform API Cost AdSense Decision WordPress.com REST API Paid plan required Paid only Cost overhead Ghost REST API Self-hosted or paid Manual setup Ops overhead Blogger v3 API Free Native integration Adopted Blogger is free, exposes a REST API, and integrates natively with AdSense. For personal blog automation, this combination i...
Read more »