Ontology and Memory Systems (6/13) — What a Good Agent Memory Architecture Looks Like

When people talk about memory, many jump straight to retrieval quality or long-term recall. Those matter. But the more foundational question is different: what should live where, when should it be loaded, and what deserves promotion into a long-term operating asset? A good memory architecture is less about storing more and more about separating storage, loading, and promotion rules.

ํ•ต์‹ฌ ์š”์•ฝ

  • Working state, session recovery, and long-term knowledge are not the same memory layer.
  • In memory systems, load rules matter earlier than storage volume.
  • Without promotion rules, memory stores become polluted quickly.
  • Without schema, provenance, and retention, memory becomes an operational contaminant instead of an operational asset.

1. What gets missed when memory architecture is treated only as retrieval quality

If memory is framed only as a retrieval problem, teams focus mostly on can the system find the right thing.

Operationally, earlier questions usually matter:

  • is this information needed right now
  • will it still matter in the next session
  • is it valuable enough to become a durable asset
  • who can correct it if it is wrong

So memory architecture is not just about search. It is first about location and lifespan.

2. Working state, session recovery, and long-term knowledge are not the same memory

Memory becomes confusing when everything is treated as one pool.

In practice, it helps to separate at least three layers:

Memory layer Example Lifespan
working state current goal, checkpoints, next action short
session recovery handoff notes, session notes, unverified items medium
long-term knowledge reusable rules, taxonomy, failure lessons, operating principles long

Without this separation, temporary work notes start behaving like permanent knowledge and durable assets disappear inside chat residue.

3. Good memory architecture prioritizes load rules before storage

It is usually more important to decide when something should enter active context than to store as much as possible.

Without clear load rules:

  • important rules get buried
  • stale context keeps following new work
  • old information starts looking current
  • context cost keeps growing

A practical load rule often asks:

  1. is this directly needed for the current goal
  2. does it influence the current decision
  3. is it old but still valid
  4. can its source and trust level be checked

That is why memory architecture is less about volume and more about activation discipline.

4. What belongs in always-visible context and what should stay outside

Not all memory should be visible all the time.

Always-visible context External memory
role, prohibitions, current goal long reference material
critical constraints for the current task older session notes
rules that must apply right now long-term knowledge stores

If that distinction is missing, memory stops being a convenience layer and becomes a context-pollution channel.

5. Why memory stores get polluted without promotion rules

One of the biggest failures in memory systems is not storage cost. It is uncontrolled promotion.

If anything can become long-term memory:

  • temporary judgments linger like permanent rules
  • incorrect information gets reused repeatedly
  • duplicate notes accumulate
  • trust in memory falls

Good promotion rules often ask:

  • will this likely be reused
  • does this apply beyond one session
  • is the source clear
  • can a human later correct or delete it

Without that filter, memory systems age badly.

6. Why schema, provenance, and retention are structural requirements

For memory to become an operating asset, three things matter early:

Element Meaning
schema what fields are stored
provenance where the memory came from
retention how long it is kept and when it expires

Without those three, the system may remember more, but it becomes harder to trust what it remembers.

7. Where memory ownership meets memory architecture

Memory ownership asks who treats memory as our asset. Memory architecture asks how is that asset structured and maintained.

They are different questions, but they connect directly.

  • without ownership, architecture can get trapped inside platform defaults
  • without architecture, ownership becomes only a declaration with no durable structure

So a strong memory system needs both the ownership boundary and the structural design.

8. What teams still need to control even when the platform provides memory

Provider memory features are convenient. Teams still need local control over:

  • which memories deserve long-term promotion
  • which memories should be deleted
  • which source metadata must be preserved
  • which load rules should govern active context

So the critical issue is not only where memory is stored. It is what rules govern how memory is operated.

9. Conclusion: good memory architecture stores less blindly and separates more clearly

Strong memory systems do not try to preserve everything forever. They draw sharper distinctions:

  • what is needed now versus later
  • one-off state versus long-term asset
  • storage versus loading
  • memory versus active context

That is why the core of memory architecture is not quantity. It is separation rules.

Related Internal Links

References

Series overview: Series index

๋Œ“๊ธ€

๋Œ“๊ธ€ ์“ฐ๊ธฐ

์ด ๋ธ”๋กœ๊ทธ์˜ ์ธ๊ธฐ ๊ฒŒ์‹œ๋ฌผ

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

← 1/10 SQLite Memory Engine f… ๐Ÿ“š Series Index 3/10 memcore Completed → A memory engine that runs without daemons, without dependencies, anywhere ํ•ต์‹ฌ ์š”์•ฝ 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"

์ด๋ฏธ์ง€
← 8/9 Deep Learning Architec… ๐Ÿ“š 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"

์ด๋ฏธ์ง€
Series overview: Series index ๐Ÿ“š 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. ํ•ต์‹ฌ ์š”์•ฝ An evaluation set for RAG is ...
Read more »

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

์ด๋ฏธ์ง€
← 7/9 Deep Learning Training ๐Ÿ“š 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. ํ•ต์‹ฌ ์š”์•ฝ CNN : weight sharin...
Read more »

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

← 6/9 Neural Networks ๐Ÿ“š 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. ํ•ต์‹ฌ ์š”์•ฝ SGD : \(w \leftarro...
Read more »

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

← 1/13 Current Structure Inve… ๐Ÿ“š 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 ๐Ÿ“š 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 »