Agent Self-Improvement Harness (12/12) — Why Router_Control Was Never Integrated

Build common infrastructure yourself vs. delegate to an external framework — a routing-layer design decision

ํ•ต์‹ฌ ์š”์•ฝ

  • Router_Control is an OpenAI-compatible proxy (/v1/chat/completions, port 3939) designed to auto-route requests across models using 14-dimension complexity scoring, targeting 70–80% cost reduction.
  • By the time the design and code were complete, the parent framework (Hermes) already provided a routing layer at the same abstraction level. The self-built proxy was never integrated and remains on HOLD.
  • What this post covers: (1) the 14-dimension complexity classification schema, (2) the architectural decision criterion "don't build common infrastructure — borrow it," and (3) how the decision not to integrate affected migration count.

Router_Control Architecture

Router_Control is structured as an OpenAI-compatible proxy. Clients only need to change OPENAI_BASE_URL.

  • Endpoint: /v1/chat/completions
  • Port: 3939
  • Behavior: receive request → score complexity across 14 dimensions → map score range to model → forward to that model
  • Target effect: 70–80% cost reduction (simpler requests downshift to cheaper models)

The key design point: classification logic and transport layer are coupled inside a single process. As discussed below, this coupling becomes a decisive variable in the integration decision.

14-Dimension Complexity Classification Schema

Each dimension is scored 0–3. The total (or weighted sum) maps to a model tier.

  1. Input token count
  2. Code / natural-language ratio
  3. Multi-language presence
  4. Request type (question / generation / transformation)
  5. Expected output length
  6. Domain specificity
  7. Estimated reasoning depth
  8. Tool-call requirement
  9. Context dependency
  10. Response-time sensitivity
  11. User label (experiment / production)
  12. Historical outcome for the same pattern
  13. Token-cache suitability
  14. Output determinism requirement

The concept is straightforward. When the sum (or weighted sum) falls below the threshold, the request goes to a low-cost model; above it, to a high-cost model. Weights are expected to be fine-tuned on production data.

Rationale for Holding Integration

At the time design was complete, two conditions held simultaneously.

  1. Parent framework has an equivalent layer. Hermes' smart_model_routing.py performs routing at the same abstraction level. The number of classification dimensions differs, but the routing layer itself is identical.
  2. Doubled migration cost. Integrating Router_Control into OpenClaw and then porting it to Hermes means transplanting the routing layer twice: first OpenClaw → Router_Control, then Router_Control → smart_model_routing.py.

Under these conditions, holding integration produces lower total porting cost. Router_Control stopped at design-and-code-complete status and remains on HOLD.

What Integration Would Have Cost

Assumption: Router_Control had been integrated into OpenClaw first.

  • 11 locations in OpenClaw would depend on Router_Control. Each dependency becomes an individual modification target during the Hermes move.
  • Moving to Hermes requires rewriting the smart_model_routing.py adapter to match the 14-dimension schema. The existing adapter cannot be reused.
  • Result: the surviving asset is not the 14-dimension classification schema but the OpenAI-compatible proxy implementation — a domain where external frameworks win decisively.

Asset Separation Principle

The generalizable form of this decision:

  • Decision logic (14-dimension classification, weights): domain-specific knowledge. An asset. Maintain as documentation and policy files.
  • Transport layer (OpenAI-compatible proxy, forwarding): generic infrastructure. Delegate to an external framework.

For Hermes integration: code porting = 0 lines, policy porting = approximately 20 lines (moving the 14-dimension weights into smart_model_routing.py's routing policy file as prompt heuristics).

Limitations and Unmeasured Items

The costs accepted by staying unintegrated are explicit.

  • The 70–80% cost reduction is a target. During the period when the self-built proxy was not running, this reduction did not materialize on the Router_Control side.
  • Hermes' smart_model_routing.py achieves a similar effect through different heuristics, but the exact reduction rate has not yet been measured (verification required).
  • Per-dimension weights in the 14-dimension schema require post-tuning on production data. The current version is an initial estimate.

Applicable Conditions

This pattern is reusable when:

  • A self-developed infrastructure component operates at the same abstraction level as an existing layer in an external framework.
  • Two porting steps need to be avoided — when the cost of the first port has a significant chance of voiding the second.
  • The decision logic and transport layer of the self-built component are separable. (If they are not separable, building it yourself may be justified in the first place.)

Open Questions

  • Of the 14 dimensions, how many actually contribute to cost reduction? If only 3–4 are effective, the schema can be simplified.
  • How much does policy ported as prompt heuristics degrade compared to code-based routing? If degradation is significant, patching the external framework's routing layer directly may be a better path.
  • Where is the boundary of "borrow common infrastructure"? How does this principle apply to cases where decision logic and transport layer are mixed and difficult to separate?

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 »