OpenClaw to Hermes Migration (3/13) — What Moves to Hermes, What Gets Dropped: 4-Category

Subtitle: The criteria for sorting 30 scripts into ✅ ๐Ÿ”ง ๐Ÿ’Ž ๐Ÿ—‘ four buckets

ํ•ต์‹ฌ ์š”์•ฝ

  • When mapping existing assets against a target platform's built-in capabilities, a Replace / Partial-Replace / Preserve / Drop four-category framework is the practical choice.
  • Hermes's core design point is the Pluggable MemoryProvider in agent/memory_provider.py — external memory is treated as a first-class citizen.
  • Applying this framework to the OpenClaw → Hermes case yields: ~15 Replace, ~12 Partial-Replace, 10 Preserve (port), 8 Drop. The Preserve bucket determines whether the migration is justified.

Classification Criteria — Why Four Categories

Agent migrations fail when treated as a binary "port everything vs. rewrite everything." Real assets are distributed across a continuous spectrum. The four categories are a simplification of a 2×2 matrix: whether the target platform exposes a matching interface × whether domain-specific logic exists.

Symbol Meaning Decision Condition
✅ Replace Target platform built-in is equivalent or superior Built-in covers the need; no domain logic required
๐Ÿ”ง Partial-Replace Interface exists; only decision logic must be ported Built-in hook/provider present; judgment criteria are domain-specific
๐Ÿ’Ž Preserve (Port) Not in target platform; unique asset No built-in; domain logic is the core value
๐Ÿ—‘ Drop Source-platform-specific; meaningless after migration Tied to source framework; cannot or should not be ported

A. Memory Layer — Separating Container from Content

Memory-related assets typically intermix "storage/retrieval mechanisms" with "judgment logic." The Pluggable MemoryProvider architecture forces these two apart.

OpenClaw Hermes Equivalent Category
bank/ 4-tier structure MemoryProvider plugin wrapping ๐Ÿ”ง
recall-tree / recall-match MemoryProvider.prefetch(query) ๐Ÿ”ง
retain-extract / retain-merge on_session_end / on_pre_compress hook ๐Ÿ”ง
memory-micro-cycle cron + sync_turn ✅
confidence-decay Not present ๐Ÿ’Ž
topics-validate / topics-expand Not present ๐Ÿ’Ž
user-pattern-stage (U-tag dialectic) Not present ๐Ÿ’Ž
bank-lint / bank-size-watch Not present ๐Ÿ’Ž
entity-audit Not present ๐Ÿ’Ž
Topic-Cued Recall (full) prefetch + system_prompt_block ๐Ÿ”ง

Principle: The container for retrieval and storage is provided by the platform. The judgment logic — what to surface, what to evict, how to decay confidence — remains a domain-specific asset. This separation is enforced by the MemoryProvider interface signature: prefetch(query) -> list[Memory] delegates storage to the provider, while selection, ranking, and decay of entries in that list belong to the implementation.

B. Model / Routing / Context — Almost Entirely Replaced

OpenClaw Hermes Equivalent Category
Router_Control proxy (port 3939) smart_model_routing.py ✅
context compaction (safeguard) context_compressor.py ✅
manual prompt cache management prompt_caching.py ✅
auth-profiles.json credential_pool.py ✅
model fallback native fallback ✅

This area is common infrastructure. Model routing, prompt caching, credential pooling, and context compaction are independent of agent domain. The only justification for keeping a custom implementation is when the built-in fails to meet a specific requirement. Otherwise, ✅ Replace is the rational choice.

C. Heartbeat / Self-Review — Half Preserved

OpenClaw Category Notes
heartbeat-router/tick/health ✅ Absorbed by cron + on_turn_start
heartbeat 11 escalation reasons ๐Ÿ’Ž Decision criteria are domain assets
self-review-prescan/apply/fix ✅ Upgraded via hermes-agent-self-evolution (DSPy+GEPA)
daily-audit Phase A (docs-snapshot) ๐Ÿ’Ž Not present in Hermes
proactive-exception-alerts ๐Ÿ’Ž 11-reason judgment preserved

Pattern: "When to check" (heartbeat timing, self-review trigger) can be delegated to the platform. Conversely, "what counts as a problem" (11 escalation reasons, docs-snapshot criteria) is domain-specific knowledge derived from operational history. Self-Review's DSPy+GEPA-based upgrade is a ✅ case — the built-in implementation is functionally superior, so there is no reason to retain the custom logic.

D. External Integrations

OpenClaw Category
scrape.py (Scrapling + playwright + curl_cffi) ๐Ÿ”ง
researcher Deep Mode (5-round AutoSearch) ๐Ÿ’Ž
gws-wrapper / gws-check ๐Ÿ”ง
Telegram Forum 6-topic routing ✅
Discord main channel ✅
oMLX integration ✅

Deep Mode's 5-round AutoSearch is the canonical ๐Ÿ’Ž example. The search protocol itself — query expansion → result collection → cross-reference → gap identification → re-search — is domain logic that does not exist in the platform.

E. Workspaces → Profiles

The 11 workspace abstractions are absorbed into Hermes's profiles feature. Structurally, roughly half of the workspaces expressed role differences as model differences, so the functional set collapses to 5–6 profiles. This is a case where a custom abstraction maps naturally onto a platform-native concept.

F. Skills

General workflow skills (brainstorming, verification, code-review, etc. — 9 total) are classified ๐Ÿ’Ž → ported to optional-skills/. They have no OpenClaw dependency and are generically reusable. The oc-* series (8 total) are ๐Ÿ—‘ — once the platform is replaced, their invocation preconditions disappear.

G. Hooks

PostToolUse (JSON validation, Linter, DOC_SYNC), PreToolUse (git commit gate), and SessionStart/End/PreCompact are all ๐Ÿ”ง Partial-Replace. They are ported to Hermes's profile/gateway hook interface, with gate logic — what passes and what is blocked — carried over intact. Hooks are a textbook case of separable "when to run" and "what to do."

Classification Summary

  • ✅ Replace: ~15 items → estimated ~30% reduction in scripts/configuration
  • ๐Ÿ”ง Partial-Replace: ~12 items
  • ๐Ÿ’Ž Preserve (Port): ~10 items — the migration's justification hinges on this bucket
  • ๐Ÿ—‘ Drop: ~8 items

If the ๐Ÿ’Ž count approaches zero, the migration is unnecessary — built-ins alone provide sufficient coverage. Conversely, if ๐Ÿ’Ž items are a majority, retaining the current platform may be more rational than switching. At 10/45 ≈ 22%, "platform replacement + core logic port" falls within the typical range that justifies migration.

Conclusion — Applicability Conditions

This classification framework is reusable under the following conditions:

  • Source and target platforms are clearly distinct (here: OpenClaw → Hermes)
  • Interface signatures on both sides can be directly compared
  • The boundary between "domain-specific logic" and "common infrastructure" is identifiable at the code level

It is a poor fit when: the abstraction levels of the two platforms differ significantly, or when the value of an asset resides not in individual code but in data (learned embeddings, accumulated interactions).

Open question: what does quantifying the porting cost of the ๐Ÿ’Ž 10 items actually yield? Measuring LOC, test coverage, and dependency count provides an upper bound on "preservation value." The next post covers that measurement.

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 »