OpenClaw to Hermes Migration (6/13) — 5-Week Plan: Feature Categorization and Profile Consolidation

A structured breakdown of schedule design, feature categorization, and profile consolidation techniques for incrementally migrating an accumulated agent platform.

What This Post Covers

  • Classification criteria for normalizing a multi-file agent system into functional categories (A–K)
  • 5-week schedule design: schema → data migration → adapters → asset porting → dual execution + cutover
  • Profile consolidation approach: selection principles for reducing fragmented workspaces
  • Daemon-to-library transition: rationale and trade-offs in a single-user system

Approach: Debt Cleanup, Not Code Port

Moving an accumulated agent system to a new runtime is not the hard part — a straight copy takes one or two days. The schedule expands when data migration, validation, and dual execution are included. The real cost of migration accumulates in data integrity and the verification window, not in the code itself.

This post defines how to divide that window and measure it.

Feature Categorization for Migration (A–K)

Viewed at the file level, the migration target is large. Viewed at the functional level, it compresses to a single-digit set of categories. The purpose of normalizing before scheduling is straightforward: schedules move by category, and categories become the unit of validation.

Category Core
A Memory System 4-tier, Retain tags, U-tag dialectic, Reflect, micro-cycle, Topic-Cued Recall, semantic search
B Heartbeat Multi-tier escalation reasons + quota checks
C Self-Review Multi-stage pipeline + Skill Evolution Loop
D Daily Audit Phases A–E (guide/update detection, diagnosis, aggregation, final review)
E Researcher Deep Mode Multi-round AutoSearch + multi-step web access
F Channel Routing Discord primary channel + Telegram Forum Topic branching
G Profile Mapping Workspace → profile consolidation
H Hooks PostToolUse, PreToolUse, SessionStart/End/Compact
I Security Model Loopback, token, dmPolicy, single-user scope
J Auto-Update com.hermes.update-check (new)
K Deprecation Targets heartbeat-tick.py, etc.

Cost Distribution

Category A (Memory System) accounts for more than half of total migration effort. The constraint is that the 4-tier structure and U-tag dialectic must be migrated losslessly while replacing the storage backend with SQLite. The 4-tier structure cannot be restored from a simple key-value dump, and dialectic tags require time-ordering and reference relationships to preserve semantic meaning.

Category K is excluded from scheduling. Explicitly inventorying assets to be discarded automatically narrows the migration scope.

5-Week Schedule

Week Work Deliverables
1 memcore SQLite schema + library skeleton Schema DDL, empty library shell
2 bank/ → memcore migration (rsync + migration scripts) Converted SQLite DB, validation report
3 Hermes install + minimal profile + MemoryProvider adapter Readable minimal runtime
4 Asset porting + Hermes profile configuration Fully functional runtime
5 Dual execution + validation + cutover Equivalence verification logs, cutover decision

Design Principle: Build Read Before Write

Weeks 1–3 focus on migrating data and making it readable. Week 4 handles logic porting. Week 5 handles equivalence comparison and cutover. Completing reads before writes surfaces migration errors early. Going in reverse order causes migration failures to accumulate until execution time, increasing rollback cost.

Dual Execution Is the Primary Safety Mechanism

Week 5's dual execution phase runs both systems simultaneously and compares output. The sequence: run the same memory micro-cycle on both sides → compare output → if equivalent, cut over; if divergent, analyze the diff. The critical point is that no switch is flipped in a single step.

This structure is a variant of the Strangler Fig pattern. The legacy system remains live while the replacement runs the same functionality in parallel. The switch is thrown only after verification evidence accumulates. Neither side is decommissioned before that threshold is reached.

Profile Consolidation

A fragmented workspace structure is the result of accumulation, not optimal design. The consolidation criterion is: do active skill categories overlap? Overlapping profiles merge; independent ones remain separate.

Profile Absorbed From Active Skill Categories
mir-coordinator mir + planner Routing, conversation, PRD
research researcher Web research, AutoSearch Deep
writer communicator Writing, editing
coder orchestrator Coding, CLI, sub-agent factory
google-workspace google Gmail, Calendar, Drive
memory-ops memory-runner + memory-manager + reviewer Bank I/O, Reflect, monthly review

What Moves Out of Profile Scope

monitor and telegram-ops are demoted from profile level to cron/hook level. Simple routing operations do not warrant a dedicated profile. The decision criterion: does the component carry state? Stateless routing is satisfied by a hook alone.

Measuring the Reduction

Reducing profile count decreases the number of switching decisions not linearly but in pair-wise proportion. The cognitive load reduction does not occur when the count is roughly halved — it occurs when each profile's role is defined as mutually exclusive.

Daemon-to-Library Transition

The most structurally significant decision in this migration is implementing the memory layer as a library, not a daemon. The legacy system ran multiple background processes — oMLX server, gateway daemon, and others. The new system handles memory exclusively through SQLite + a Python library (memcore).

Hidden Costs of a Daemon

In a single-user system, a daemon imposes the following requirements:

  • Process death detection — a mechanism to notice when the process has died
  • Restart policy — choosing among launchd, systemd, or supervisor
  • Health checks — endpoint design + periodic invocation
  • Log rotation — preventing disk saturation
  • Concurrency control — required when multiple clients are assumed

A library requires only import. In a single-user environment, concurrency problems do not arise. SQLite's WAL mode satisfies single-writer + multi-reader access requirements.

Transition Scope

oMLX is retained — embedding serving is not easily replaced by a library. The gateway and memory daemons are fully absorbed into the library layer. As a result, a significant number of LaunchAgents are scheduled for removal.

Limitations and Open Questions

This plan applies to single-user systems. In a multi-user environment, the daemon-to-library transition does not hold — concurrency and resource isolation become required. Profile consolidation is similarly valid only when roles can be defined as mutually exclusive.

Open questions:

  • During dual execution, what magnitude of output divergence qualifies as "equivalent"? Byte-level comparison is infeasible when LLM sampling introduces nondeterminism. A semantic equivalence criterion is needed.
  • Category A's dialectic tags require preserving a reference graph. Whether to model this as a table structure in SQLite or as a JSON column remains an open decision.
  • Can profile count be reduced further? Usage patterns may support merging mir-coordinator and writer.

Where This Applies

The schedule design and categorization method in this post are reusable in the following scenarios:

  • Migrating an accumulated agent system to a different runtime
  • Migrations that involve schema replacement alongside logic porting
  • Cases where workspace or profile count has fragmented beyond actual need
  • Determining when a daemon-based architecture can be absorbed into a library

The core principle: include the verification window explicitly in the schedule. When only code porting is scheduled, verification becomes "something to do when done" — and done never arrives.

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 »