A precise explainer of native tool calling (function calling): what the protocol is, the request/response JSON on the wire, how the model emits it, and why small models fail to produce it.

An LLM only ever outputs text — so how does an AI agent run a tool? The answer is the native tool call, a precise, trained, parseable protocol that most people hand-wave over. Here is exactly what it is, what happens on the wire, and why a small model that knows what to do still cannot make one.

What a Native Tool Call Actually Is

An empirical study of small-model coding agents on a 4 GB RTX 3050: edit format, model choice, retry, decomposition, role-split, and a custom text tool-call loop, measured across ~900 automated trials.

We stopped theorizing and ran ~900 trials on a 4 GB laptop GPU to measure what makes a small local model a good coding agent. The result overturns intuition: the edit format and the editor model decide almost everything, the clever agentic scaffolding mostly hurts, and the simplest setup won.

The 4 GB Experiment: What Actually Makes a Small Model a Good Coding Agent

A micro-level look at how AI coding agents write, run, test, verify and debug code: the edit loop, output acceptance, the exec tool, and why some agents feel like native command-line programmers while others feel like scripters.

The first four posts in this series took the telescope view of AI agents. This one takes the microscope: how a coding agent turns an instruction into a working, tested change — how it edits, compiles, runs, tests and debugs — and why Claude Code behaves like a native *nix programmer while others reach for a throwaway script.

How a Coding Agent Actually Writes Code

A strengths-and-weaknesses comparison of the open coding agents Codex, Pi, OpenCode, Kimi Code and Gemini CLI, plus why Claude Code arguably has the best harness despite being closed.

After reading the source of five open coding agents — OpenAI's Codex, Pi, OpenCode, Kimi Code and Gemini CLI — here is how they actually stack up, strengths and weaknesses, which one wins, and why the best harness of all belongs to a tool we deliberately left out of the code study: Claude Code.

The Coding Agent Shootout and the Best Harness We Could Not Read

How AI agents capture knowledge through skills and memory, why humans still write some skills, and what self-improving means in code — based on the source of six open agent harnesses.

Self-improving AI agents sound like marketing — until you read the code. This is how agents capture knowledge as skills and memory, why a human still has to write some of it down, and what self-improving actually means once you strip away the hype: bookkeeping, an LLM review pass, and a human gate.

Skills and Memory: How AI Agents Actually Learn

A deep dive into the AI agent harness — context management, tool contracts, safety boundaries, persistence and recovery — based on reading the source code of six open agent harnesses.

We read the source code of six AI agent harnesses — Pi, Hermes, OpenCode, Kimi Code, OpenAI's Codex CLI and Google's Gemini CLI. This is what the harness really does for the model, why almost every agent failure is a harness failure, and why the engineering lives there, not in the model.

What an AI Agent Harness Actually Does

A code-level comparison of OpenAI Codex, Pi (pi.dev) and Nous Research's Hermes Agent, using their differences to explain the different types of AI agents.

Comparing OpenAI Codex, Pi (pi.dev) and Hermes Agent by Nous Research — an odd trio at first glance, but their differences map out the whole landscape of AI agents: what each type is, what it can do, and which one can help you.

Codex, Pi and Hermes: Three Agents, Three Species

Tynion v2 turns a tag-picking prompt builder into a visual glossary for AI image generation. Every lighting style, camera angle, composition rule, and pose now ships with a definition and an AI-generated example image — same base subject, one variable changed — so you can see what a term does before you put it in your prompt.

Tynion, our text-to-image prompt helper, just got its biggest update yet: nearly 1,000 prompt terms, and every one of them now explains itself. Hover over any term — Golden Hour, Rule of Thirds, Contrapposto — and a card pops up with a plain-English definition and a real example image showing exactly what that term does to a picture.

Tynion v2: See What Every Prompt Term Looks Like

A practical walkthrough of deploying FileBrowser Quantum as an HTTP gateway over SMB network shares, issuing scoped long-lived JWT tokens, and wiring the API into Claude Desktop via a local MCP server — so AI agents can access office files without VPN or physical LAN access.

Claude Cowork (and similar browser-based AI agents) can browse web pages but cannot reach SMB/CIFS network shares. Here is how we wrapped our office NAS in an HTTP API so the agent can list, read, search, and write files from anywhere — with proper security controls.

Giving AI Agents Access to Network Drives via an HTTP File Server

A hands-on walkthrough of every error you'll encounter setting up Google Sign-In with Strapi 4 behind an nginx reverse proxy — invalid callback URLs, secure cookie failures, redirect_uri_mismatch, email conflicts, and role bugs.

Strapi's built-in Google OAuth sounds straightforward until you actually try it. Here's every error I hit and exactly how I fixed them.

Configure Strapi To Use Google Login: Not a Trivial Task

Testing Anthropic Messages API model IDs including claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, and claude-sonnet-4-6 returned HTTP 200 with matching model metadata, but some assistant self-descriptions claimed Qwen or DeepSeek. Here is the raw observation and why model self-identification should be treated cautiously.

Raw Anthropic API tests showed model metadata matching Claude model IDs, while some Chinese self-identification prompts produced Qwen or DeepSeek answers.

Why Does Claude Opus 4.8 Say It Is QWen

Using The Bravados and a real legal dispute involving ChatGPT, this essay looks at AI hallucinations, factual verification, and the difference between using AI as a sparring partner and treating it as an oracle.

A confident wrong answer about The Bravados becomes a lesson in when to trust AI, when to verify it, and why plausible answers can be dangerous.

Who Killed Mrs. Douglas

Learn how to build a faster, cheaper alternative to traditional VPNs using Google Cloud Run's serverless architecture for web proxy deployment.

Ditch complex VPN setups for a lightweight serverless web proxy using Google Cloud Run

Serverless Web Proxy on Cloud Run: A Better VPN?

A practical note on fixing a Cloud Run custom domain that stays stuck on 'Waiting for certificate provisioning' behind Cloudflare. Explains why orange-cloud proxying breaks validation and how switching the record to DNS-only lets the SSL certificate complete.

Cloud Run domain mapping can stay stuck on certificate provisioning when Cloudflare proxying blocks Google's SSL validation.

Cloud Run Domain Mappings with Cloudflare: Why Your SSL is Stuck

A hands-on benchmark of 13 local LLMs — Llama, Qwen3, DeepSeek-R1, Coder models and more — running on a home server with 64GB RAM and RTX 3060 12GB VRAM via llama.cpp. Includes GSM8K math, MMLU knowledge, and HumanEval coding scores, real token speeds, and an honest account of what doesn't fit.

We benchmarked 13 local LLMs on a machine with 64GB RAM and an RTX 3060 12GB using llama.cpp router mode — here's what scored well, what failed to load, and what surprised us.

13 Local LLMs, One RTX 3060: What Actually Runs (and What Doesn't)

Moving beyond UI generation: using AI to debug, integrate, and ship a real trading profit/loss calculator from prototype to production

How AI helped me turn an AI-generated frontend mockup into a fully functional trading calculator

AI Beyond the UI: From v0 Prototype to a Finished Product

A step-by-step guide to using WetraceTool to extract your WeChat database encryption key on Windows, so you can read your own message history with tools like decrypt_sync.

WetraceTool is a small Windows utility that walks you through extracting your WeChat SQLCipher database encryption key in four guided steps — no command line required.

How to Extract Your WeChat Encryption Key with WetraceTool

A practical guide to setting up Cygwin SSHD on Windows and using Claude Code to manage it remotely — far beyond what browser-based AI assistants can do.

Stop fighting with Windows GUIs. With Cygwin, an SSH daemon, and Claude Code, you can manage any Windows machine from the command line — uninstalling software, cleaning the registry, killing startup entries — just by asking.

Managing Windows Made Easy with Cygwin and Claude

I spent time with Claude CoWork, Anthropic's answer to OpenClaw and GUI-based AI automation. Here are my honest thoughts on its browser-sandbox limitations, where it genuinely shines, and why I still reach for Claude Code for the heavy lifting.

A first-hand look at Claude CoWork — Anthropic's GUI-based AI agent — exploring its browser-only scope, how it compares to Claude Code, and what the road ahead might look like for AI-driven desktop automation.

My Thoughts on Using Claude CoWork: The GUI Frontier

A practical walkthrough of configuring a Cisco 1941/K9 router as a home office gateway backup. Covers console access from Alpine Linux using screen, pulling config from a live router via SSHv1 in Docker, adding IPv6 (SLAAC + DHCPv6), pushing config over serial, and fixing policy routing on a dual-NIC Linux host.

How to set up a Cisco 1941/K9 as a home office gateway — including console access via serial, cloning config from a running router, adding IPv6 with SLAAC and DHCPv6, and fixing dual-NIC asymmetric routing on Alpine Linux.

Configure a Cisco 1941 Router as Your Home Office Gateway with IPv6

A step-by-step guide to installing Alpine Linux on old, low-powered hardware and configuring it as a mobile backup gateway using Android USB tethering. Covers BIOS quirks, disk identification, NAT/iptables setup, and hotplug automation with mdev.

Turn an old, dusty PC into a reliable 4G/5G backup internet gateway using Alpine Linux and Android USB tethering — no expensive hardware required.

Repurpose Your Old PCs: How To Install Alpine Linux on Them

A step-by-step guide to setting up an Alpine Linux box as a backup WAN gateway for your home or small business network, using Android USB tethering as the mobile data source. Includes a failover script and full network configuration.

Turn an old PC or mini box running Alpine Linux into a backup WAN gateway using your existing Android phone's USB tethering — no extra SIM, no extra plan, no extra cost.

Create a Backup Mobile Data WAN for Your Small Business / Home Network

Discover which forex brokers genuinely support cTrader FIX API access for retail traders - spoiler alert: it's not as many as you'd think.

A reality check on which forex brokers actually allow retail traders to use cTrader's FIX API

cTrader FIX API: Which Brokers Say Yes to Retail Traders?

When our office Meraki router died, we scrambled through Ubiquiti replacements, PPPoE struggles, and a sneaky VLAN configuration that taught us valuable lessons.

A real-world network disaster story: router failure, emergency replacement, and the VLAN gotcha that almost broke everything

Router Down! My 3-Day Battle with Meraki, Ubiquiti & VLANs

Struggling with RDP showing a black 'Please Wait' screen? Learn how pending Windows updates cause this issue and get the simple fix.

Fix RDP black screen with 'Please Wait' message by updating Windows - a simple solution that works.

RDP 'Please Wait' Black Screen? A Simple Fix

Navigate China and Japan's digital payment systems and electric vehicle landscape with practical tips for tech professionals traveling abroad

Technical insights on digital payments and EV adoption across China and Japan for tech travelers

China & Japan at a Glance: Cashless Futures and Electric Cars

Learn about the React2Shell vulnerability (CVE-2025-66478) affecting Next.js apps, how attackers exploit it, and immediate steps to secure your applications.

Critical CVE-2025-66478 vulnerability allows remote code execution in Next.js applications worldwide.

React2Shell: Critical RCE Flaw in Next.js & How to Fix It

A detailed journey from VirtualBox chaos to a stable Windows 11 VM on KVM/QEMU, with AI assistance turning complex command-line wizardry into a manageable migration.

When VirtualBox kept freezing my Windows 11 VM, I turned to KVM/QEMU. The problem? Nobody remembers all those command-line incantations.

Escaping VirtualBox Hell: How I Migrated Windows 11 to KVM with AI

A technical case study showing how AI can effectively diagnose mechanical problems, turning a frustrating mower issue into a quick fix.

How AI helped diagnose a violent mower shake caused by a tiny wood chip, saving time and guesswork.

AI Found the Simple Fix for My Shaking Mower

Learn how AI can assist with practical home problems by walking through a real troubleshooting session that fixed a shaky washing machine.

How AI helped diagnose and fix a vibrating washing machine through step-by-step troubleshooting.

AI Fixed My Shaky Washer: A Troubleshooting Guide

Debugging Linux suspend issues can be frustrating. This article chronicles the systematic troubleshooting of immediate wake-from-suspend problems, exploring USB controllers, sleep modes, and ultimately discovering a Realtek Ethernet controller as the culprit.

A deep dive into troubleshooting Linux suspend issues, from USB controllers to sleep modes, and how AI transformed a potentially frustrating debugging session into a systematic investigation.

How to put your baby, sorry, your Linux box to sleep

Discover how Google Gemini, Codex, and Claude AI collaborated to build Koala-Tinue, a Chrome extension that keeps CommSec sessions alive automatically.

How AI tools created a Chrome extension that solves CommSec session timeouts with zero human coding.

This Chrome Extension for CommSec Was Coded Entirely by AI

Comprehensive guide to resolving .NET Core build hangs and library dependency errors on Linux using Docker containerization for consistent builds.

Learn how Docker containers solve persistent .NET build hanging issues on Linux distributions.

Solving the .NET Build Hang on Linux with Docker

Complete guide to fixing the 'could harm your computer' preview error in Windows File Explorer for PDFs from network drives and internet sources.

Resolve Windows File Explorer preview errors for PDF files from network shares and internet downloads.

Fix the 'Could Harm Your Computer' File Preview Error

Compare Google Gemini, Claude Code, and OpenAI Codex from a developer's perspective. Learn which AI coding assistant fits your needs and budget.

A developer's honest take on Google Gemini, Claude Code, and OpenAI Codex for programming tasks.

AI Coder Showdown: Gemini vs. Claude vs. Codex

A detailed rebuttal to the MakeUseOf article about switching from Linux to Windows, explaining why the author's reasoning reveals fundamental misunderstandings about Linux capabilities

A software developer's response to common misconceptions about Linux, addressing claims about dual-booting, virtual machines, and the real power of Unix-like systems

Why That 'I Ditched Linux for Windows' Article Misses the Point Entirely

A real-world example of how AI assistants like Claude Code are revolutionizing technical troubleshooting, turning tedious configuration problems into quick wins.

How I used Claude Code to instantly fix a frustrating Thunderbird browser configuration issue that would have taken hours to troubleshoot manually.

From Hours to Seconds: AI-Powered Troubleshooting Thunderbird Browser Issue with Claude Code

Troubleshooting checklist and final registry and group policy changes to let a MetaTrader 5 gateway reconnect to a legacy SMB share on Windows 11.

Steps I used to restore MetaTrader 5 access to an SMB share after Windows 11 hardened guest authentication.

Fix: Cannot Connect to SMB Share on Windows 11

When mathematical formulas refuse to render on a web page, which AI coding assistant can actually solve the problem? I tested Google Gemini, OpenAI Codex, and Claude Code on the same MathJax bug. The results surprised me.

A real-world comparison of Google Gemini, OpenAI Codex, and Claude Code in debugging a MathJax rendering issue. Spoiler: only one AI actually fixed it.

From Broken Fractions to Beautiful Formulas: Claude Code Trumpeted the Game!

A practical story of using AI-assisted development to fix open source software bugs, featuring axios compatibility issues and path resolution problems in the Cloudflare CLI tool.

How Claude Code helped me debug and fix critical bugs in cloudflare-cli, transforming a broken tool into a reliable DNS management solution in minutes.

Empowering Open Source Development with AI: My Experience Fixing cloudflare-cli

This detailed walkthrough explains how we rebuilt a legacy MODX website on Next.js, upgraded Bootstrap 3 to 5, restored missing content, modernized the contact form, and partnered with Claude Code, Gemini CLI, and OpenAI Codex to automate the migration.

Case study of migrating a MODX site to Next.js with AI copilots guiding Bootstrap upgrades, content recovery, and automation.

Migrating a Website from MODX to Next.js with the Help of AI Tools

Building a PC with an MSI PRO Z790-P motherboard? You might face a no-boot issue if both CPU power cables are connected. This guide explains why you should start with just one cable (CPU_PWR2) to avoid troubleshooting and only add the second for extreme overclocking.

Building a PC with an MSI PRO Z790-P motherboard? You might face a no-boot issue if both CPU power cables are connected.

Read This Before Connecting Power Cables to the MSI PRO Z790-P Series Motherboard

In this article, I will share my experience of using AI to transcribe and analyze audio files.

Let the Audio Speak: Using AI to Decode and Analyze Conversations at Scale

In this article, we explore the essential sources of EOD data and demonstrate how to store and access it efficiently using Redis, a high-performance in-memory database.

Building Blocks of Algorithmic Trading: Collecting End-of-Day Data

In this article, we explore how AI can help you diagnose car issues and save money on mechanic fees. Learn how to use AI tools like ChatGPT and Claude to troubleshoot car problems before visiting a mechanic.

In this article, we explore how AI can help you diagnose car issues and save money on mechanic fees.

Using AI as Your Personal, Cost-Free Mechanic

In the fast-paced world of trading, information is power—but only if you can access and synthesize it effectively. Over the years, I’ve collected an extensive library of trading materials: research reports, technical guides, strategy notes, economic models, and annotated charts. The challenge has always been how to organize this knowledge and use it dynamically. Onyx, a LLM Chat UI that can be used for building Retrieval-Augmented Generation (RAG) systems

In this article, I’ll walk through how I used Onyx to turn a trading knowledge base into a responsive AI assistant.

Finding Your Edge: Build a RAG System with Onyx for Better Trading Strategies

In this post, I share my experience using v0.dev to create a frontend for my CLI tool that calculates profit and loss for trading. I discuss the tools I used, the project goals, and the steps I took to build a user-friendly interface.

In this post, I share my experience using v0.dev to create a frontend for my CLI tool that calculates profit and loss for trading. 

Jumpstarting My Project UI with v0.dev: From CLI Tool to Web App

In the world of finance, words can be as powerful as numbers. This article explores the vocabulary that can trigger market reactions and how traders can use this to their advantage.

In this article, we explore how certain words can trigger market reactions and how traders can leverage this knowledge for better decision-making.

Take Some Words Seriously: Adding Edge to Your Trading

In this article, we delve into the words that can cause market volatility, from commodities to currencies to equities. Understanding this vocabulary is essential for traders and analysts.

In the world of finance, words can be as powerful as numbers. This article explores the vocabulary that can trigger market reactions.

Words That Move Markets: The Most Dangerous Vocabulary in Finance

In this post, we explore the hidden vocabulary of the market and how it can be used to gain an edge in trading. We discuss the importance of understanding the language of the market and how it can impact trading decisions.

In this post, we explore the hidden vocabulary of the market and how it can be used to gain an edge in trading.

The Hidden Vocabulary of the Market

This guide demonstrates how to leverage tyo-crawler to convert Evernote notes into Markdown, creating a well-organized knowledge base that can be used with open-source RAG tools for enhanced AI applications.

Learn how to transform your Evernote notes into a structured Markdown knowledge base using tyo-crawler, making them ready for integration with Retrieval-Augmented Generation (RAG) systems.

Building an AI Knowledge Base: Using tyo-crawler to Convert Evernote Notes into Markdown

In this article, we will explore how to use tyo-crawler to scrape tweets from X.

How to Scrape Twitter (X) User's Tweets Using tyo-crawler

In this blog post, we explore `tyo-crawler`, a Node.js web crawler built on Puppeteer. Learn how to harness its features for efficient web scraping.

Unleash the Power of Puppeteer: A Deep Dive into the TYO-Crawler Web Crawler

Discover the power of `tyo-crawler`, a Node.js web crawler built on Puppeteer. Learn how to harness its features for efficient web scraping.

Open-source tools provide a range of features for building a Retrieval-Augmented Generation (RAG) system, enhancing large language models (LLMs) with external knowledge retrieval. In this post, we explore some of the most prominent open-source tools for implementing RAG-based applications, including Open Web UI, Verba, Onyx, LobeChat, RagFlow, RAG Web UI, Kotaemon, and Cognita.

In this post, we explore some of the most prominent open-source tools for building a Retrieval-Augmented Generation (RAG) system, including Open Web UI, Verba, Onyx, LobeChat, RagFlow, RAG Web UI, Kotaemon, and Cognita.

Building a Retrieval-Augmented Generation System Using Open-Source Tools

In this article, we'll suggest building an affordable AI machine with great scalability and over 256GB of memory. We'll cover different build tiers, component choices, and considerations to help you achieve a powerful AI setup without overspending.

In this guide, we'll suggest building a scalable AI machine with over 256GB of memory while keeping costs manageable.

Building an Affordable AI Machine with Great Scalability (256GB+ Memory)

In this guide, we explore the importance of memory capacity in AI workloads and provide recommendations for building your own AI rig with more memory and power. We break down AI machine builds into three categories based on memory capacity: 64GB–192GB, >256GB, and >512GB, and recommend components for each category.

Building an AI rig with more memory and power can significantly improve performance for AI workloads.

Building Your Own AI Rig: More Memory, More Power

In this post, we explore the best free and open-source chat UIs for LLMs, including Open Web UI, LibreChat, AnythingLLM, ChatGPT-UI, GPT4All, oobabooga, Verba, Dify, SillyTavern, Danswer, Lobe UI, Hugging Face Chat-UI, KoboldCpp, PrivateGPT, Serge Chat, Jan, etc.

In this post, we explore the best free and open-source chat UIs for LLMs, including Open Web UI, LibreChat, AnythingLLM, etc.

Exploring the Best Free and Open-Source Chat UIs for LLMs

In this post, we explore two new open-source LLM chat tools: Cherry Studio and ChatBox, highlighting their unique features and LLM integrations.

Empowering Your Ideas with AI: Exploring New LLM Chat Tools

In this article, we'll explore the various document processing tools available and how to choose the right one for your specific workflow needs. Whether you're looking for open-source PDF-to-Markdown conversion, AI-powered data mining, lightweight parsing, or advanced AI-powered document structuring, we'll provide insights into each tool's strengths and weaknesses.

Choosing the right document processing tool depends on your specific workflow needs. Here's a guide to help you find the right tool for your needs.

Finding the Right Document Processing Tool for Your Workflow

This article provides a comprehensive list of Linux (Ubuntu) commands for video processing, including tools for converting, editing, and transcoding videos. It covers various tasks such as resizing, cropping, and rotating videos, as well as using ffmpeg for advanced features like audio mixing and video effects.

In this post, I will show you a list of handy Linux (Ubuntu) commands for video processing.

Handy Linux (Ubuntu) Commands for Video Processing

This article explores six AI video generation tools—**HailuoAI, KlingAI, Luma, Pika Labs, Runway, and Sora**—to present their capabilities, strengths, and potential use cases. 

In this post, I will show you what KlingAI, Runway, Pika Labs, Luma, Sora, and HailuoAI offer in the field of AI video generation.

Exploring AI Video Generation: What KlingAI, Runway, Pika Labs, Luma, Sora, and HailuoAI Offer

In this post, I will show you how to fix an autoplay issue in React/NextJs.

Fixing an Autoplay issue in React/NextJs

In this article, we will cover the steps required to create celebrity look-alike images with Stable Diffusion, including prompt control, fine-tuning, and advanced techniques for enhancing results.

In this tutorial, we will explore the steps required to create celebrity look-alike images with Stable Diffusion.

Creating Celebrity Look-Alike Images with Stable Diffusion

In this article, we will cover the steps required to master human portraits with Stable Diffusion AI, including model selection, prompt engineering, and advanced techniques for enhancing results.

In this tutorial, we will explore the steps required to master human portraits with Stable Diffusion AI.

Mastering Human Portraits with Stable Diffusion AI

In this article, we compare four popular AI-powered tools - Stable Diffusion, Ideogram, FlexClip, and ChatGPT - to see which one delivers the best results in generating human portraits.

Creating a Human Portrait with AI: Which Model Does It Best?

In this guide, we explore the key parameters of Stable Diffusion, an AI image generation model, and how to fine-tune them for optimal results.

Stable Diffusion Parameter Guide

In this article, I will guide you through installing, configuring, and using Stable Diffusion WebUI Docker to streamline your image generation workflow.

Using Stable Diffusion WebUI Docker for Image Generation – A Newbie's Guide

In this post, I will show you how to use math equations and LaTex expressions in MDX documents with `next-mdx-remote`.

Using Math Equations and LaTex Expressions in MDX Documents with Next.js

This article will show you how to generate images with AI models on your own computer.

How to Generate Images with AI Models on Your Own Computer

In this post, we will test the performance of the deepseek-r1 model with Ollama. We will run the test with 32B parameters and explore the model's performance in math, coding, and reasoning.

In this post, we will test the performance of the deepseek-r1 model with Ollama.

Testing Deepseek-r1 Model with Ollama

This article focuses on one of the most common issues: unexpected costs in cloud computing. By examining real-life cautionary tales, we’ll explore why these costs occur and how you can avoid them.

This article focuses on one of the most common issues: unexpected costs in cloud computing. 

TYO LabWe Build

Latest