AI Field Notes

Real AI experiments with screenshots, prompts, and practical verdicts.

This is the hands-on side of zglg.work: model tests, coding agents, local LLM setups, tool comparisons, and workflow experiments written for global AI readers. Each note keeps the evidence visible so you can judge the result instead of only reading a summary.

Read latest field note Explore AI tools

From Field Evidence to Buyer Research

Use hands-on AI tests to choose software, APIs, agents, and security tools

Buyer research

AI Software Buyer Guides

Use the field notes as evidence, then compare AI software by workflow, controls, integration cost, and buyer risk.

Open research path

Model choice

AI Model Benchmarks

When an experiment depends on model quality, check benchmark sources, scenario rankings, latency, price, and transparency.

Open research path

Developer tools

Best AI Coding Agents

Turn coding-agent field notes into a shortlist for Codex, Claude Code, Cursor, Windsurf, and repo-aware workflows.

Open research path

Enterprise search

Enterprise AI Search Tools

Move from document experiments to enterprise search, knowledge management, RAG controls, and workplace retrieval software.

Open research path

API decision

OpenAI vs Anthropic API

Use test evidence to compare model APIs by cost, reliability, tool use, context, safety, and vendor fit.

Open research path

Security

LLM Security Tools

When experiments touch private data or agents, compare guardrails, red teaming, prompt injection, and AI security tooling.

Open research path

AI Field Notes FAQ

Turn AI experiments into tool research decisions

How should I use AI field notes before buying software?

Use each field note as practical evidence, then compare the related software category, model benchmark, API guide, security checklist, and workflow cost before choosing a tool.

Are AI news posts enough to choose an AI tool?

No. AI news and screenshots help you spot useful capabilities, but production decisions still need vendor comparisons, pricing checks, privacy review, integration testing, and fallback planning.

Which AI field notes are most useful for overseas readers?

Start with hands-on tests that include screenshots, prompts, coding-agent workflows, model comparisons, local LLM setups, and tool verdicts you can verify against your own workflow.

What should I read after an AI experiment looks promising?

Open the related buyer guide, AI model benchmark page, coding-agent comparison, API provider guide, enterprise search guide, or LLM security comparison before spending time or budget.

Latest field note2026-07-06#125

Qwen3.5 9B + Hermes: The Best Local Agent Setup I Have Tested So Far

A hands-on test of Qwen3.5 9B with Hermes, covering local inference speed, Ollama setup, memory behavior, image understanding, OCR, Excel analysis, and why this may...

Open field note

Experiment Library

AI tests you can actually inspect

113 field notes online

2026-06-07 · #112

Gemma 4 12B Local Agent Test: Running a Multimodal Model on a Laptop

A hands-on Gemma 4 12B test with Ollama and DeepLocals, covering local deployment, inference speed, RAG over local files, Word export with citations, and multimodal...

2026-06-05 · #111

MiniMax M3 Coding Benchmark: Testing It Against GPT-5.5 and DeepSeek-V4

A hands-on MiniMax M3 benchmark across three coding-agent tasks: an Excel analytics web app, a Three.js smart factory scene, and a screenshot-to-HTML UI recreation,...

2026-06-03 · #110

DeepSeek-V4 and MinerU for Scanned PDF RAG: A Practical Local Knowledge Base Test

A screenshot-rich test of using MinerU with DeepSeek-V4 inside DeepLocals to parse scanned PDFs, preserve formulas, retrieve local documents, and combine private fil...

2026-06-02 · #109

I Finally Found Free Tokens for MyClaw: Qwen3.6-35B Works Directly (Tutorial)

A tutorial-style MyClaw field note showing how Qwen3.6-35B can run through iFlytek SparkDesk MaaS free API quota for coding and data-analysis workflows.

2026-05-31 · #108

I Tested Qwen3.7-Max Against Claude Opus 4.8 and DeepSeek-V4. The Result Surprised Me.

A Qwen3.7-Max benchmark against Claude Opus 4.8 and DeepSeek-V4 on an AI agent task, with screenshots and practical ranking notes.

2026-05-28 · #107

I Tested a Chinese AI Agent: In 20 Minutes It Produced Word, PPT, Charts, and a Full Report

A Tiangong AI agent field note showing a 20-minute workflow that generates Word documents, PPT outlines, charts, and a full report from one request.

2026-05-27 · #106

I Tested a Local AI Agent for One-Person Company Workflows

A one-person company workflow test using a local AI agent to read files, collect web data, analyze information, and produce business-ready documents.

2026-05-26 · #105

I Tested a Local Knowledge Base That Learns Hundreds of Files and Exports Reports

A local knowledge-base test showing how an AI assistant learns hundreds of private files, retrieves evidence, answers questions, and exports summary reports.

I Built an AI Agent Team: From Finding Trends to Producing a Video

2026-05-25 · #104

Real AI experiments with screenshots, prompts, and practical verdicts.

Use hands-on AI tests to choose software, APIs, agents, and security tools

AI Software Buyer Guides

AI Model Benchmarks

Best AI Coding Agents

Enterprise AI Search Tools

OpenAI vs Anthropic API

LLM Security Tools

Turn AI experiments into tool research decisions

Qwen3.5 9B + Hermes: The Best Local Agent Setup I Have Tested So Far

AI tests you can actually inspect

Gemma 4 12B Local Agent Test: Running a Multimodal Model on a Laptop

MiniMax M3 Coding Benchmark: Testing It Against GPT-5.5 and DeepSeek-V4

DeepSeek-V4 and MinerU for Scanned PDF RAG: A Practical Local Knowledge Base Test

I Finally Found Free Tokens for MyClaw: Qwen3.6-35B Works Directly (Tutorial)

I Tested Qwen3.7-Max Against Claude Opus 4.8 and DeepSeek-V4. The Result Surprised Me.

I Tested a Chinese AI Agent: In 20 Minutes It Produced Word, PPT, Charts, and a Full Report

I Tested a Local AI Agent for One-Person Company Workflows

I Tested a Local Knowledge Base That Learns Hundreds of Files and Exports Reports

I Built an AI Agent Team: From Finding Trends to Producing a Video

I Tested Gemini-3.5 Against DeepSeek-V4 and GPT-5.5. The Result Was Unexpected

DeepSeekMine Is Now DeepLocals: V3.1.2 Improves Local File Learning Speed

I Connected DeepSeek-V4 to Hermes: Crawling Dozens of Web Pages with One Sentence

An AI That Understands My Files: Learning Thousands of Local Documents and Generating Reports

A Chinese Alternative to Claude Code: Open Source, No Configuration, Ready to Use

DeepSeek-V4 with a Layout-Preserving Translation Agent Is Surprisingly Practical

Run Qwen3.5 9B GGUF Locally with LM Studio: DeepSeek-V4 Local Deployment Tutorial

RHTV Video Generation Agent Test: Editable Steps and Transparent Workflow

DeepSeek-V4 with Claude Code: A Hands-On Coding Agent Test

Stop Chasing Prompts: Why the Harness Decides AI Coding Success

Qwen3.6 27B Test: Can a 27B Model Challenge GPT-5?

DeepSeek-V4, GLM-5.1, and GPT-5.5 Compared Again: Practical Results

DeepSeek-V4 vs GLM-5.1 vs GPT-5.5: Model Comparison Test

Next-Generation Image Model Test: When Reality Starts to Look Synthetic

AI Learns 2,000 Web Pages and Generates a Word Summary

Hermes Agent Tutorial: Install and Use Skills That Accumulate Automatically

Local File AI That Exports Word Reports: DeepLocals Workflow Test

Free Tokens for OpenClaw Workflows: A Practical Setup Note

AI-Built Full-Book PDF Translation App: Preserve Images and Export Word

AI That Learns 10,000 Computer Files Across 30 Formats

OpenClaw Visual Guide: A Clear PDF Tutorial

OpenClaw Excel Data Collection Agent: Automate Work Without Mouse Clicks

OpenClaw Knowledge Base Tutorial: AI Q&A and Content Monetization

Secure Your Computer After Installing OpenClaw: AI Safety Checklist

Alibaba Qwen Model Takes First Place Again: Practical Model Notes

OpenClaw Safe Edition: One-Click Install and Direct Use

Local File Learning AI That Writes Deep Summary Reports

Local Knowledge Base Software for Mac and Windows: Beginner Guide

AI Agent Generates Data Analysis Results from One Sentence with Free Gemini-3

Alibaba's Latest Model Takes First Place Again: Qwen Release Notes

AI That Learns 10,000 Local Files with 4 Million Characters per File

An AI Agent That Can Control Both Phone and Computer

Install OpenClaw Out of the Box: Beginner-Friendly Setup

A More Reliable AI Knowledge Base for the Holiday Season

OpenClaw Local Deployment: Let an Agent Operate Your Computer Files

OpenClaw Deployment from Zero: 30-Page Beginner Tutorial PDF

AI Agent Self-Evolution Test: How an Agent Learned to Improve Itself

The Easiest OpenClaw Installation Method: Two Commands

Install OpenClaw Locally on Your Computer: Beginner-Friendly Method

OpenClaw One-Click Installer for Windows and Mac

OpenClaw WeChat Integration: Enterprise WeChat Customer Service Bot

Connect Personal WeChat to OpenClaw: Build a Website from a Phone Message

OpenClaw Feishu Integration: 24/7 A-Share Market Monitoring Assistant

OpenClaw Skill Tutorial: Generate Word Documents from a Fixed Template

An AI That Learns Local Computer Files and Writes Deep Reports

HAIC 2025 and the New Solution for Expensive AI Agent Compute

Yuandong Tian's Five-Year PhD Reflection: Reading Notes

Open-Source Voice AI with Near-Zero-Latency Conversation

AI Talking Avatar Tutorial: Generate Script, Voice, Subtitles, and Video

Open-Source AI Agent Workflow Generator: Build Workflows from One Sentence

Deploy Qwen3-Coder on a Personal Computer: Top Coding Model Tutorial

MiniMax Agent Builds a Jay Chou Song Website in One Click

Why DeepSeek, GPT-4o, Qwen, and Kimi Answered That OpenAI's CEO Was Tim Cook

AI Agent for Full PDF Book Summaries: Export Word and PPT Reports

DeepSeekMine Local Knowledge Base: Docker Version Preview

Gemini Screenshot-to-HTML and Sketch-to-Game Test

Docker Packaging and Distribution: Complete Practical Guide

Qwen for Scanned PDF to Word: Full-Book Document Conversion Test

Build an AI Agent That Publishes to Xiaohongshu with One Click

Learn AI Agents from Scratch: Beginner-Friendly Agent Tutorial