Guides
Why AgentKits Memory's Hybrid Search Has to Solve Japanese Twice
How AgentKits Memory fuses BM25 and vector search, and why Japanese, Chinese, and Korean queries need a second decision in the keyword half.
agentkitssearchrag
AityTech Blog
Guides, news, and tutorials from AityTech on manufacturing AI, factory DX, legacy code modernization and software development.
Guides
Most COBOL Modernization Tools Stop at the Program. The Batch Job That Calls It Is Where Migrations Actually Break.
Why Legacy Dragon parses JCL, with condition codes, PROC nesting, and GDG windows, as a first-class language in the same AST.
legacy-dragonjclmainframe
Guides
322 Voices, 142 Languages, No Server: What It Takes to Fit That Much Speech in a Browser Tab
Why voice and language breadth, not speed, is the hard problem for on-device browser TTS, and how PrivateAI's 322+ voices compare.
privateaittson-device
Guides
AgentKits Marketing Has 28 Skills. It Never Loads More Than Five.
How the AgentKits Marketing Kit's selector loads at most five of its 28 skills to avoid context rot, and why it is simpler than the research proposal.
agentkitscontext-engineeringai-agents
Guides
Japan's ¥12 Trillion Legacy-System Warning Isn't a COBOL-Skills Story. It's an Encoding Story.
Why Legacy Dragon treats EBCDIC and Shift-JIS character encoding as a parsing-layer concern, not a pre-processing step added later.
legacy-dragoncobolmainframe
Guides
No Pricing Page, No API Meter: The Economics Behind PrivateAI's Free Tools
Why cloud AI is priced per token while on-device tools have no such bill, and what that buys, and does not, for PrivateAI.
privateaion-devicewebgpu
Guides
Two Plugins Live Inside AgentKits Marketing. Its Marketplace Listing Shows One.
Why the agentkits-marketing repo holds two plugins, Content Factory and Campaign Manager, while marketplace.json lists only one, and what it signals.
agentkitsplugin-architecturemarketplace
Guides
The Engineer Kit Has More Skills Than the Marketing Kit and Still Isn't Live
Why AgentKits' Engineer Kit (40 skills, Preparing) was built after the Marketing Kit (18 agents, 93 commands, 28 skills, Ready).
agentkitsengineer-kitroadmap
Guides
RBS Locked Out 6.5 Million Customers Over One Batch Update. The Failure Wasn't the Change — It Was Not Knowing What It Touched.
Why RBS's 2012 meltdown and TSB's 2018 migration failed on unseen impact, and why Legacy Dragon's AST graph is built for impact analysis.
legacy-dragoncobolmainframe
Guides
Chrome Now Ships Its Own On-Device AI. Where Does That Leave Tools Like PrivateAI?
What Chrome 148's stable Summarizer, Translator, and Language Detector APIs on Gemini Nano cover, and where PrivateAI fills the gap.
privateaiwebgpuon-device
Guides
The Translator Is an Agent Too: How AgentKits Localizes Its Own Docs
How AgentKits ships its README and 23-module course in ten languages with a Claude Agent SDK tool, and why an LLM suits the job.
agentkitslocalizationclaude-agent-sdk
Guides
The Generative AI Rewrite Pitch Is Everywhere in Legacy Modernization. Legacy Dragon Deliberately Isn't One.
Why generative COBOL-to-Java rewrites are not reproducible, and why Legacy Dragon starts from a deterministic AST graph instead.
legacy-dragoncobolgenerative-ai
Guides
410 Million Reasons Not to Upload: PrivateAI and the Shadow AI Problem
Why PrivateAI's browser-only architecture sidesteps the shadow AI leakage behind ChatGPT's 410 million DLP violations in 2025.
privateaiprivacyshadow-ai
Guides
COBOL Gets the Headlines. VB6 and PowerBuilder Are the Legacy Crisis Nobody's Watching.
Why Legacy Dragon parses VB6 and PowerBuilder with the same AST-based approach it uses for mainframe COBOL, the quieter legacy estate.
legacy-dragonvb6powerbuilder
Guides
One .claude Folder, Eighteen Installer Targets: How AgentKits Handles AI-Tool Config Fragmentation
How AgentKits' installer supports 18 AI coding tools from one .claude/ source, and how it relates to the AGENTS.md standard.
agentkitsopen-sourceclaude-code
Guides
The Real Deadline for COBOL Modernization Isn't the Code — It's the People Who Understand It
Why Legacy Dragon's AST graph matters for capturing the knowledge of retiring COBOL engineers as well as for migration speed.
legacy-dragoncobolmainframe
Guides
AgentKits Memory Bets on One Local SQLite File While the Agent-Memory Market Bets on the Cloud
Why AgentKits Memory stays in the local-MCP lane among Mem0, Zep, Letta, Cognee, and LangMem, and what that trade-off costs and buys.
agentkitsmcpai-agents
Guides
The Compliance Case for AI That Never Touches a Server
How WebGPU makes browser-based AI fast and 2026 data-localization law makes it valuable, and why on-device tools like PrivateAI fit.
privateaiprivacywebgpu
Guides
Why Legacy Dragon Ships as a Single Binary: Parsing COBOL Where the Network Doesn't Reach
Why Legacy Dragon ships as a single binary for COBOL and JCL shops with tight network controls, with no npm install or cloud API call.
legacy-dragoncobolmainframe
Guides
Why AgentKits Organizes Agents by Funnel Stage, Not by Tech Stack
Why AgentKits' Marketing Kit groups agents by TOFU/MOFU/BOFU instead of technical domain, a deliberate bet on depth over breadth.
agentkitsopen-sourceclaude-code
Guides
Why Zunavi Is On-Premise First: Semiconductor Drawings Don't Leave the Building
Why semiconductor and precision manufacturers cannot use cloud drawing SaaS, and how Zunavi makes on-premise the default architecture.
zunavimanufacturingon-premise
Guides
Certain vs Optional: How Zunavi Decides What to Trust
How Zunavi's feature flags separate reliable capabilities from assistive ones that need human verification, and why that is enforced twice.
zunavimanufacturingfeature-flags
Guides
Why We're Rebuilding Zunavi's Drawing Ledger From Scratch
Why tracking the latest drawing revision by deleting and reinserting rows breaks, and how PostgreSQL, Prisma, and one flag replace it.
zunavimanufacturingpostgresql
Guides
AI-Assisted Parsing for Legacy Code Modernization
Why parsing is the bottleneck in legacy modernization, and how AST-based tools for COBOL, JCL, and PL/I change the economics.
legacy-modernizationcobolast
Guides
The Numbers Behind Shipping Six Products: How We Actually Use Claude Code
What AityTech's git history shows about its product portfolio: conventional commits, an agent-driven workflow, and where AI tooling pays off.
engineeringclaude-codeworkflow
Guides
Inside PrivateAI: 15+ AI Tools That Never Leave Your Browser
How PrivateAI's browser-based OCR, text-to-speech, speech-to-text, background removal, and translation tools run without a server.
privateaiprivacybrowser
News
Introducing Zunavi: A Drawing Management Platform for Manufacturers (In Development)
Zunavi (図ナビ) is our in-development, on-premise-first platform for managing manufacturing drawings, specs, and approvals.
zunavimanufacturingdrawing-management
News
Introducing the AityTech Portfolio: Five Products, One Philosophy
A look at the AityTech lineup of PrivateAI, Legacy Dragon, AgentKits, Yomite, and StudyKits and its privacy-first, on-device philosophy.
aitytechportfolioprivacy
News
Welcome to the AityTech Blog
Our blog: guides, news, and tutorials on iOS development, web platforms, AI solutions, and open-source tools from our studio in Japan.
announcementaitytechblog
Tutorials
Building Privacy-First AI Apps on macOS
How we architect macOS applications that process AI workloads entirely on-device — no cloud uploads, no data collection, no compromise on capability.
macosprivacyai