Ontology and Memory Systems (3/13) — What Changes When You Ontologize Yourself

OpenClaw 비용 관리와 실전 운영 노하우 — 3개월 운영 회고

Starting from the fatigue of re-explaining yourself to AI every single session.

Key Takeaways

  • If the agent ontology (Parts 1–2) is the system's map, the personal ontology is the user's own map.
  • Structure: Being → Values → Capabilities → Actions — four layers.
  • Eliminates repetitive self-explanation to AI; accelerates fit-evaluation for new opportunities.
  • A résumé is the self you present to others. An ontology is the structural diagram you maintain for yourself.

Background

Part 1 defined the ontology governing relationships between agents. Part 2 covered its implementation.

One element was missing. The most critical information shared across agents is not the system architecture — it is the user.

Multiple agents all work for the same person. If that person's identity, decision criteria, and priorities are not structured, every new session starts from scratch with a full self-introduction.

That repetitive inefficiency is where personal ontology begins.

모델 비용 3계층(Tier) 아키텍처

Body

Reflect 파이프라인의 구조적 비용 절감

1. What Is a Personal Ontology

The agent ontology answered: "Which agents exist in the system, and how are they connected?"

The personal ontology answers: "Who is the person operating this system?"

It is not a profile. "Name, occupation, interests" gives AI almost nothing useful.

What is needed is the structure of judgment — why you do this work, what criteria drive your choices, what capabilities you hold, and what you are doing right now. The 4-layer framework organizes exactly that.

2. The 4-Layer Framework: Being → Values → Capabilities → Actions

L0. Being — What kind of person am I?

The foundational layer. Rarely changes.

What belongs here: - Disposition: where you draw energy, how you process change, what sustains focus - Non-negotiables: things you will not compromise regardless of circumstance - Reaction patterns: how you behave under stress, how you reorient after failure

This layer tells AI: "This is fundamentally who this user is." Regardless of the project or the agent, this disposition remains constant.

L1. Values — What criteria govern my decisions?

The priority stack for judgment.

What belongs here: - Value ranking: your priority ordering (e.g., growth vs. stability, autonomy vs. efficiency) - Decision habits: do you concede first or counter first? intuition or data? - Emotion-judgment coupling: do your decisions shift with mood, or remain stable?

A critical observation belongs here: declared values and behavioral values often diverge. You say "growth is my top priority," yet your actual choices consistently favor stability. Recording that gap is the core value of an ontology.

L2. Capabilities — What can I do?

Current skills, meta-skills, and — critically — gaps.

What belongs here: - Expert domains: what you can execute at depth - Intermediate domains: what you can do, but are not expert-level - Gaps: what you cannot do or lack

Explicitly documenting gaps matters most. When AI knows the user's capability boundary, it calibrates explanation depth and produces appropriate recommendations. Listing only strengths yields a résumé, not an ontology.

L3. Actions — What am I doing right now?

The highest-churn layer.

What belongs here: - Active projects - Short- and medium-term plans - Routines and habits

3. Two Directional Flows the 4 Layers Enable

This structure is not a flat list. It enables two directions of reasoning.

Top-down tracing: "Why?"

"Why am I running this project?" → Which capabilities does it engage? (L2) → Which values underlie that choice? (L1) → Which aspect of my being does it originate from? (L0)

When direction is lost, traversing upward re-anchors the root motivation.

Bottom-up prediction: "What's next?"

Being (disposition) + Values (decision criteria) + Capabilities (what is possible) combined → → predicts what actions are most likely to follow.

When a new opportunity arrives, filtering it through this framework answers quickly: "Interesting, but misaligned with my values" or "A capability gap, but consistent with my being-layer." Decisions that previously required lengthy deliberation resolve in seconds.

4. Why a Hierarchy, Not a Profile

"Why not just write a detailed self-introduction?"

A profile is enumeration. An ontology is hierarchy. The distinction is concrete.

A profile reads: "Flutter developer, interested in investing, growth-oriented." AI treats these three items as equivalent, flat data. It cannot determine which takes precedence, why you chose Flutter, or how investing and development connect.

The 4-layer framework makes the structure legible: - L0 (Being): interest-driven focus; exits when growth stalls - L1 (Values): growth > autonomy > efficiency - L2 (Capabilities): embedded systems expert, Flutter intermediate, investing beginner - L3 (Actions): app development, blog, investment analysis

With this structure, AI reads different information: "This user pursues Flutter because of growth (L1). If interest erodes, relocation to another domain is likely (L0)." The depth of subsequent recommendations changes qualitatively.

Hierarchy also makes conflicts visible. "Growth is my top priority" (L1) while "I repeatedly choose the stable option" (L3) — this tension is not a bug. It maps precisely to the actual friction in current life. A profile never surfaces this. A hierarchy makes it automatic.

5. Applied to AI Agents

Summarize this personal ontology in a global CLAUDE.md. Every agent in the system then recognizes the user's identity without being told.

Four concrete effects:

① No more repeated self-explanation

New session, new agent — no manual "I'm the kind of person who..." The global CLAUDE.md carries that context.

② Agents act in context

A development agent that knows "this user wants the counter-argument first" leads with alternatives. A content agent that knows "data-based reasoning is preferred" sources evidence before framing. Same user understood, applied through each agent's specific role.

③ Inter-layer conflicts surface the actual tensions in current life

Mapping all four layers reveals conflicts between them: "growth is my top priority" (Values) versus "I keep making stable choices" (Actions). This is not a malfunction — it precisely matches what the user is actually experiencing. The ontology becomes an instrument of self-understanding.

④ Ontology vs. résumé

A résumé is the self curated for external presentation — capability- and experience-focused, shaped for the audience.

An ontology is a structural diagram maintained for oneself — disposition, values, and gaps included, exactly as they are. What you expose to AI should be the structural diagram, not the curated résumé. Processed input produces processed output.

Caveats

  • Completeness on day one is not required. Start at L3 (Actions). Repeatedly ask "why am I doing this?" and work upward. The upper layers fill in naturally.

  • The gap between declared and actual values must be recorded honestly. Writing an idealized self produces idealized recommendations. Writing the actual self produces actionable ones.

  • An ontology is not static. L0 (Being) is nearly stable. L1 (Values) shifts through significant experiences. L2 (Capabilities) updates through learning. L3 (Actions) changes continuously. Periodic revision is required.

Closing

Part 1 defined the relationships between agents. Part 2 covered the implementation. Part 3 addresses the structural definition of the user.

Ontology is ultimately three layers: - Agent ontology: who is in the system and how they connect - Communication ontology: how data flows and knowledge is layered - Personal ontology: who the person operating this system is

When all three layers are in place, AI agents become not tools, but a system that understands you. Whether you operate a few agents or many, the starting point is the same — define yourself first.

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 »