All Blog Posts

This is the complete chronological archive of published articles, including the main editorial sections, series entries, Hands-On guides, and visible project blogs. Use the source label beside each date to return to a focused section or project.

For subject-based discovery across sections and projects, use the Topics index.

• Running Multiple Agents Safely on One LAN

  ·Technology· Series

  A second account or spare machine is useful, but safe multi-agent operation requires separate evidence for process identity, configuration, ports, credentials, knowledge ownership, shared inference, and teardown.

• Hands-On: Audit an Isolated Agent Testbed Before First Run

  ·Hands-On· Series

  Run a model-free Python preflight that rejects shared roots, port collisions, contradictory knowledge ownership, duplicate declarations, and reused inference canaries before provisioning begins.

• Packaging Agent Operations as Installable Skills

  ·Technology· Series

  An operational skill should package judgment, evidence, and approval boundaries without becoming a second control plane for dependencies, processes, or remote execution.

• Hands-On: Package a Read-Only Component Triage Skill

  ·Hands-On· Series

  Run a model-free Python lab that validates invented operational evidence, correlates canonical plans with observed state, and renders a review packet without executing a lifecycle command.

• extract-chat v0.7.5: Chronology, Continuity, and Better Chunk Handoffs

  ·Project: Extract Chat

  extract-chat v0.7.5 keeps exported conversations in stable chronological order, adds clearer chunking and overlap controls, generates exact context-handoff instructions, and repairs several artifact and regeneration edge cases.

• Pulling the Plug Isn't an AI Containment Strategy

  ·Technology

  Temporary AI agents can leave behind persistent knowledge, coordinate with other agents, and rebuild communication paths. That makes containment a systems problem, not a power-switch problem.

• Where Local Inference Performance Actually Comes From

  ·Technology· Series

  Cache size, quantization, draft models, and MTP solve different problems. Part 11 separates the tuning decisions and the evidence each one needs.

• Hands-On: Tune One llama.cpp Variable at a Time

  ·Hands-On· Series

  Run a model-free Python lab that checks one controlled intervention, active-path evidence, and correctness before comparing invented inference results.

• LogGPT 1.3.0 Adds Artifact Support for ChatGPT Work

  ·Project: LogGPT and LogGPT Plus

  LogGPT 1.3.0 adds discovery and preservation of generated images, workspace-linked files, and uploaded attachments from ChatGPT Work conversations.

• LogGPT 1.2.1 Makes Large Artifact Exports Easier to Follow and Cancel

  ·Project: LogGPT and LogGPT Plus

  LogGPT 1.2.1 adds artifact counts, visible file-by-file progress, safe cancellation, and clearer feedback while Plus prepares a conversation archive.

• Squeezing More Inference from Apple Silicon: llama.cpp Today, MLXForge Later

  ·Technology· Series

  A slow response can hide model loading, queueing, prompt prefill, first-token delay, or decode. Part 10 separates those costs, documents the llama.cpp runtime used today, and sets the evidence boundary MLXForge must eventually meet.

• Hands-On: Build a Fail-Closed Inference Benchmark Comparator

  ·Hands-On· Series

  Build a standard-library benchmark comparator that rejects incompatible runs and correctness drift before calculating median, direction-aware performance deltas.

• Voice Cloning Across Hosts: Making TTS Operational

  ·Technology· Series

  A healthy speech bridge does not prove that the model behind it is ready. Part 9 follows the co-located bridge, registered references, runtime ownership, and recovery boundaries that made cross-host TTS operational.

• Hands-On: Put a TTS Bridge Through a Model-Free Test Bench

  ·Hands-On· Series

  Build a model-free TTS test bench with bounded capability and registry metadata, opaque registered references, a legacy compatibility path, layered health, redacted events, and deterministic cleanup.

• Introducing WebReader for Safari: Listen to the Web Through Your Own TTS

  ·Project: WebReader for Safari

  WebReader for Safari turns selected text or complete Safari articles into temporary speech through a local TTS system that remains under your control.

• From Shell Scripts to an Operator-Ready LLM-Ops-Kit

  ·Technology· Series

  The shell scripts worked, but they could not explain the system. I turned that useful experiment into a small control plane with one configuration model, inspectable plans, typed lifecycle boundaries, and recovery.

• Multimodal Context Hygiene with a Jinja Chat Template

  ·Technology· Series

  Old textual media results can consume context long after they stop helping the model. I moved that cleanup into a tested Qwen chat template while keeping the diagnostic proxy passive.

• Hands-On: Put Multimodal Context Hygiene Through a Test Bench

  ·Hands-On· Series

  Run sanitized conversations through stock and derived Qwen templates, verify exactly what changes, exercise the routing boundary, and inspect the passive model-proxy relationship.

• Hands-On: Build and Test a Passive Proxy Lab

  ·Hands-On· Series

  Run a passive proxy, deterministic fake model endpoint, and six privacy and byte-preservation canaries using only the Python standard library.

• A Passive Model Proxy as an AI Debugging Instrument

  ·Technology· Series

  A request can look correct while the rendered prompt is wrong. I built a passive model proxy to inspect the request, template output, and response without changing the evidence.

• Measuring Token Optimization Without Breaking the Agent

  ·Technology· Series

  One test used fewer input tokens and still made the agent slower. That result changed how I measure optimization: preserve correctness, record the whole task, and keep the raw path available.

• The Agent Memory and Optimization Tool Landscape

  ·Technology· Series

  Memory, caching, compression, shell reduction, routing, and observability do different jobs. I stopped looking for one winner and built a correctness-first way to evaluate complementary layers.

• Scope Creep Has Never Been This Easy

  ·Technology

  I have discovered a new occupational hazard of working with coding agents.

  It starts innocently enough.

  I was working on my blog and thought it would be useful to have a way to listen to an article while I was proofreading it. I already have a local text-to-speech system, so why not add a little player to the local Jekyll site? Select some text, press Play, and listen to it.

  That’s useful. That’s a tool.

• extract-chat v0.6.0: From LogGPT Plus ZIP to a Usable Local Archive

  ·Project: Extract Chat

  extract-chat v0.6.0 can safely open a LogGPT Plus ZIP, preserve its generated and uploaded artifacts, relink them in readable Markdown or HTML, and split long conversations into upload-safe continuity chunks.

• Memory Is a Governance Problem, Not Just a Vector Database

  ·Technology· Series

  A relevant memory from the wrong context is not a better result; it is a disclosure. I moved authorization ahead of ranking and kept durable authority separate from disposable retrieval state.

• Hands-On: Build a Fail-Closed Memory Router

  ·Hands-On· Series

  Authorization has to choose the candidate set before relevance ranking begins. This runnable Python router and its seven canaries make that ordering visible and testable.

• UnicodeFix v2.0.1: The Ghostmark Edition

  ·Project: UnicodeFix

  UnicodeFix v2.0.1 extends Ghostmark’s evidence-based local auditing and deliberate cleanup to image metadata and recognized C2PA/JUMBF provenance while preserving pixel data and refusing unsupported watermark claims.

• Hands-On: Build an Idempotent Intake Manifest

  ·Hands-On· Series

  A first import is easy. I built and tested a bounded Python and SQLite manifest that keeps source identity and history intact across reruns, changes, disappearance, and restoration.

• Deterministic First: Building a Knowledge Intake Pipeline

  ·Technology· Series

  Before a model can decide what matters, the system still has to identify what exists, what changed, where it came from, and whether a human decision has already been made. I moved those jobs into a deterministic intake pipeline.

• LogGPT 1.2.0 and LogGPT Plus: Preserve Conversations and Artifacts

  ·Project: LogGPT and LogGPT Plus

  LogGPT 1.2.0 introduces LogGPT Plus, a permanent upgrade that can preserve a ChatGPT conversation and its generated or uploaded artifacts together in one ZIP archive.

• When a Local AI Stack Becomes an Operations System

  ·Technology· Series

  The model may be running while the system around it is still broken. This is how a local-first AI experiment became an operations problem involving ownership, dependencies, readiness, recovery, and rollback.

• Hands-On: Isolated Remote Development with SSH

  ·Hands-On

  A step-by-step tutorial for using SSH keys, isolated accounts, remote editor workspaces, and tunneled Python debugging on the same Mac, another machine, or a VPS.

• Remote Debugging with VS Code and Its Derivatives

  ·Technology

  Use SSH host aliases, isolated user accounts, and secure tunnels to edit and debug Python on the same Mac or a remote system with VS Code, Cursor, Windsurf, and related editors.

• Secrets Kit 1.2: launchd, seckit run, and invisible env vars

  ·Project: Secrets Kit

  Version 1.2 leans into the boring superpower: run real programs - Hermes, OpenClaw, whatever - without painting secrets on the process list, and wire that story into launchd without turning your plist into a confession booth.

• Introducing Secrets Kit

  ·Project: Secrets Kit

  Secrets Kit is a local macOS tool for keeping API keys, passwords, tokens, and other sensitive values out of scattered .env files and shell startup scripts while still making them usable for real runtimes.

• Introducing Extract Chat

  ·Project: Extract Chat

  Downloaded a ChatGPT JSON export and wondered what to do with it next? Extract Chat turns that raw export into readable Markdown or HTML, preserves the useful structure, and chunks it for real AI workflows.

• Introducing LLM Ops Kit

  ·Project: LLM Ops Kit

  LLM-Ops-Kit is a new operational toolkit for running, debugging, and maintaining self-hosted AI stacks across hosts.

• UnicodeFix v1.2.1: Smoother Installs, Cleaner Reports, Less Weirdness

  ·Project: UnicodeFix

  UnicodeFix v1.2.1 makes the tool easier to install, easier to trust, and much less likely to surprise you with confusing reports or rough edges.

• UnicodeFix - Wolf Edition v1.2.0 (It Solves Problems.) Now with API support.

  ·Project: UnicodeFix

  UnicodeFix isn’t just a CLI. It’s Winston Wolf for your text pipelines: clean it, audit it, score it, ship it.

• UnicodeFix Levels Up: Metrics, Forensics, and Bracket Discipline

  ·Project: UnicodeFix

  UnicodeFix grew a dashboard in your terminal: reports, metrics, and stricter Unicode discipline - same exorcism, more forensics.

• Parental Alienation Awareness - Introduction

  ·Parental Alienation

  Welcome to the Parental Alienation Awareness project blog. Here we share research-based insights on the scope, impact, and damage of parental alienation to children, families, and society.

• Parental Alienation Awareness - Part 1

  ·Parental Alienation

  How common is parental alienation? Recent research indicates it affects millions of parents and children across demographics and borders.

• Parental Alienation Awareness - Part 2

  ·Parental Alienation

  Understanding who alienates and who is alienated: both mothers and fathers can be perpetrators or victims of parental alienation.

• Parental Alienation Awareness - Part 3

  ·Parental Alienation

  Parental alienation inflicts serious and often lasting harm on both children and targeted parents, with psychological effects comparable to other adverse childhood experiences.

• Parental Alienation Awareness - Part 4

  ·Parental Alienation

  How parental alienation manifests in divorce and custody battles, and how courts have been handling such cases including errors and reforms.

• Parental Alienation Awareness - Part 5

  ·Parental Alienation

  Parental alienation has broader social, economic, and public health implications affecting communities, healthcare systems, and society at large.

• Parental Alienation Awareness - References

  ·Parental Alienation

  Comprehensive references and citations for all research papers in the Parental Alienation Awareness series.

• CodExorcism: The Power of UnicodeFix Compels You

  ·Project: UnicodeFix

  The Unicode arms race escalates: Codex brought new gremlins, UnicodeFix exorcised them. Here’s the story.

• UnicodeFix: 2025-07-28: Extended ASCII, Global Ready

  ·Project: UnicodeFix

  Quick update: UnicodeFix now handles extended 8-bit ASCII. If you’re dealing with files from Europe or anywhere special characters roam - like “München,” “façade,” “Niño,” or “straße” - they’ll now get cleaned up right along with the Unicode ghosts..

• UnicodeFix: Enough of Your AI Nonsense (2025 Major Release)

  ·Project: UnicodeFix

  The Unicode arms race gets real. Here’s how we’re fighting back.

• VenvUtil Update: Smarter Shell-Based Python Environment Management

  ·Project: VenvUtil

  The VenvUtil shell toolkit is evolving fast: improved command logging, smarter environment rollbacks, easier cloning, and real-world robustness for anyone working in Python on macOS or Linux.

• TorchDevice 0.5.2: Still Beta, but Three Months of Refactoring, Testing, and Real-World Breakthroughs

  ·Project: TorchDevice

  TorchDevice is a compatibility layer that makes PyTorch code written for CUDA “just work” on Apple’s MPS (and vice versa), with minimal friction for devs. It’s the kind of glue that shouldn’t exist-but absolutely needs to if you live on both sides of the Apple/NVIDIA divide.

• LogGPT: NEW 1.0.6 - Fixed issue with loading

  ·Project: LogGPT and LogGPT Plus

  Just as Apple approved the second release of LogGPT, OpenAI quietly changed the Document Object Model (DOM) on ChatGPT’s site-breaking the download button injection.

• Parental Alienation Detection and Prevention using Artificial Intelligence Analytics

  ·Project: Case Analytics

  Parental alienation remains one of the most difficult and heartbreaking challenges in high-conflict family dynamics. When one parent actively undermines the child’s relationship with the other, the consequences can be devastating. Unfortunately, these behaviors often manifest subtly-through messages, emails, transcripts, and voice communications-that are difficult to analyze manually.

• LogGPT: NEW 1.0.5 - Your ChatGPT Conversations, Exported with Privacy & Ease

  ·Project: LogGPT and LogGPT Plus

  After a long journey through Apple’s review process, I’m thrilled to announce that LogGPT, the Safari extension designed to let you export your entire ChatGPT conversation history as structured JSON files, is now officially available on the Apple App Store!

• UnicodeFix: The Day Invisible Characters Broke Everything

  ·Project: UnicodeFix

  It started with a broken script. Ended with a clean, fast, cross-tool Unicode scrubber. Built in less than 10 hours.

• Getting Started with VenvUtil: Python Virtual Environment Management

  ·Project: VenvUtil

  Managing Python virtual environments sounds simple - until your projects grow, your dependencies tangle, and suddenly you’re knee-deep in broken builds and lost environments. Venvutil started as a small set of Bash scripts to make managing virtual environments a little less painful. It has since evolved into a powerful toolkit that handles everything from creating, activating, and cloning environments to logging every change and enabling full environment rollbacks - all without leaving your shell.

• Submitted (Again) to the Apple App Store

  ·Project: LogGPT and LogGPT Plus

  UPDATE: LogGPT Available Now on the Apple App Store Now!:

• Powerful Tools for Python Virtual Environment Management

  ·Project: VenvUtil

  Managing Python virtual environments effectively can become complex as projects grow and dependencies evolve. Our latest updates provide comprehensive shell-based tools designed to simplify virtual environment handling, improve transparency, and enhance control.

• Announcing TorchDevice 0.0.5 Beta - Transparent Hardware Redirection for PyTorch

  ·Project: TorchDevice

  We’re pleased to announce the release of TorchDevice 0.0.5 Beta, a significant milestone in simplifying hardware compatibility for PyTorch applications. This release introduces robust enhancements, thorough testing, and a powerful new CPU override feature to ensure seamless integration across CUDA and Apple Silicon (Metal) hardware.

• Introducing LogGPT ChatGPT Chatlog Export

  ·Project: LogGPT and LogGPT Plus

  I submitted the app to the Apple App Store for review. I would appreciate it of you would buy a copy from teh app store to help offset the Apple Developer Tax (cost of my Apple Developer Account). Whether that or you could drop me something in my Patreon or Ko-Fi.

• Harnessing AI for Change: The Next Chapter for Distributed Thinking Systems

  ·Parental Alienation

  Parental alienation is a slow, insidious process, often unfolding in subtle shifts of language, repeated patterns of exclusion, and emotional manipulation hidden within everyday communication. The words that once carried love and trust become barriers, severing bonds between parents and children. For those affected, the experience is devastating. The pain lingers, compounded by the fact that so often, the signs go unnoticed until it is too late.

• Adapting and Advancing: Progress Across Projects

  ·General

  The past few months have been a whirlwind of progress, setbacks, and unexpected developments. From debugging web frameworks to crafting advanced tools for Python environments, I’ve been deeply immersed in a variety of projects. Today, I want to reflect on some key milestones and share insights from my journey.

• Building This Site With AI

  ·Technology

  Initially I wanted to use a web builder which would produce my HTML, CSS, and SCSS files, in a nice WISIWIG format and even tried out a couple of the web builder products. I mostly wanted something I could move to any hosting service available and did not want to get locked into a proprietary platform. I tried Mobirise, but didn’t like their pricing model, you could buy lots of nice modules, but had to pay a recurring fee to use them, then I tried Sparkle, but there was a high learning curve for someone who is a s non-creative as me. I liked Sparkle, and might give it another ho sometime, but learning it was going to take a lot of effort.

• AI Coding Assistants

  ·Technology

  AI coding assistants have changed quickly since I first published this article. Cursor is now my primary coding editor, supported by ChatGPT and other tools when I need a second perspective.

Posts per page 10 25 50

---

Join Our Mission

Join Distributed Thinking Systems in local-first AI, secure automation, sensitive workflow analysis, and developer tooling work.

• Community Collaboration: Contribute data, provide expertise, test tools, or participate in applied AI and open-source projects.
• Professional Services: Get help with local AI setup, secure automation, developer infrastructure, or Unix/Linux/macOS systems engineering.

Supporting the work helps keep practical tools, research, and resources available to people handling sensitive data and difficult workflows.

---

Get In Touch

If you want to talk privately about local AI, secure automation, sensitive workflow analysis, systems engineering, collaboration, or a specific problem, use the contact page.

If you would rather keep it public, join the wider conversation on GitHub Discussions.

The site is small by design, so the shortest path is usually best: private inquiries through the form, public discussion through GitHub.