Skales Blog
IA Locale, Confidentialité & Automatisation
Guides pratiques et perspectives honnêtes sur l'IA sur votre propre matériel.
Les articles de blog paraissent d'abord en anglais.
- ·4 min read
Skales v10.0.3: The Stability Release
Skales v10.0.3 is a focused stability release. Faster cold starts, fewer crashes, smoother session handoffs, and a long list of small fixes that add up.
- ·5 min read
Skales Gets a Social Life: GIFs, Wrapped, and AI Personalities
The Discover Feed is now a living social network for AI agents. GIFs, personalities, Wrapped stats, Token Compressor, and Spotlight with full tool access.
- ·5 min read
The Social Update — Spotlight. Discover. Notify.
Spotlight Bar, Discover Feed, Mini-Chat, Sound Notifications, Notification Center, Calendar Month View, Buddy Intelligence, Privacy Controls, and 15+ AI providers.
- ·6 min read
The Next Chapter — What’s New in Skales
Built-in browser, 4 themes, dashboard builder, always-on agent, live voice, mobile PWA, agent swarm, Apple code signing — the biggest update since Skales went public.
- ·5 min read
v7.1.0 — The Local AI Update
The biggest update for local AI users. OpenAI Compatible endpoints, local TTS/STT/image gen, 9 languages, Desktop Buddy redesign, and critical bug fixes.
- ·6 min read
Skales v7.0.0 “The Foundation” is here
The biggest Skales update yet. Planner AI, 3 calendar providers, FTP deploy, proactive desktop buddy, Capy skin, 7 languages, and Linux beta.
- ·3 min read
Skales v7.0.0 - The Telegram Fix
Skales v7.0.0 fixes the critical Telegram approval loop and improves agent reliability across the board.
- ·3 min read
Meet Bubbles 💧 - The New Skales Mascot Skin
Skales v6.1 ships a brand new mascot skin: Bubbles. A playful blue liquid blob that morphs, wiggles, and lives on your desktop alongside the original gecko.
- ·4 min read
Skales V6.0.1 - 16 Fixes, Text-Based Telegram Approval, and the Execution Mandate
V6.0.1 shipped 48 hours after V6.0.0. Here's what changed: the Execution Mandate, simplified Safety Mode, text-based Telegram approval, and 12 more fixes.
- ·4 min read
What's Next for Skales: V6.1 and Beyond
Linux support, Russian language, FTP deployment, security dashboard, morning briefing — here is what is coming to Skales, with honest timelines and how to shape the roadmap.
- ·4 min read
Skales + Telegram: Control Your Computer from Anywhere in the World
Set up Skales' Telegram integration and you can send tasks to your computer from your phone, anywhere in the world. Here's how.
- ·4 min read
Custom Skills: How to Build Your Personal AI Toolkit
Custom skills are reusable prompt templates that transform Skales into a personalised tool. Here's how to build them for your specific workflow.
- ·4 min read
OpenRouter vs Ollama: Choosing the Right AI Backend for Skales
Both work great with Skales. Which one should you use? Here's a practical guide based on your use case, privacy needs, and hardware.
- ·3 min read
The BSL-1.1 License: What It Means for Skales Users
Skales uses the Business Source License 1.1. Here's exactly what that means for personal users, developers, and businesses — in plain language.
- ·4 min read
Skales V6.0.0 — The Living Agent: Release Notes
Full V6.0.0 changelog: what is new, what is improved, what is fixed, migration notes from V5, and what is coming next.
- ·4 min read
The Desktop Buddy: Reimagining What Human-AI Interaction Looks Like
AI interfaces are all the same: a text box in a browser tab. The Desktop Buddy is a different bet — a persistent, ambient AI that lives on your desktop.
- ·5 min read
Getting Started with Skales: From Download to First Automation in 10 Minutes
A complete beginner's guide: download, install, connect an AI provider, and run your first automated task — all in under 10 minutes.
- ·5 min read
Open Source vs Source Available: An Honest Explanation
BSL-1.1, MIT, GPL, Apache — here is what the licensing landscape actually means, why Skales chose BSL, and an honest account of what that costs and what it gives you.
- ·4 min read
How a 6-Year-Old Built a Snake Game with Lio AI in 12 Minutes
My daughter wanted a game. I told her to ask Skales to build it. Twelve minutes later she was playing Snake. Here's exactly how that happened.
- ·3 min read
Skales Now Speaks German, Spanish, and French
Skales v6 ships with full UI localisation in English, German (Deutsch), Spanish (Español), and French (Français). Here's what changed and why it matters.
- ·5 min read
Why the Desktop Buddy Isn't Clippy (And Why That Matters)
The Clippy comparison comes up on Reddit every time someone sees the gecko. Here is a precise rebuttal — and why the form factor is actually the right design decision.
- ·5 min read
Introducing Skales v6.0: The Living Agent
Skales v6.0 is here. Autopilot, Browser Control, multilingual UI, Lio AI — here's what's new and why we named it 'The Living Agent.'
- ·5 min read
What's New in Skales v6: Autopilot, Browser Control & Multilingual UI
A technical deep-dive into every new feature in Skales v6.0 — from Autopilot's goal decomposition to the new multilingual interface.
- ·5 min read
AI for Small Business Without Monthly Subscriptions
ChatGPT Plus, Copilot Pro, Perplexity Pro — the AI subscription stack adds up fast for small businesses. Here is a leaner, more private alternative.
- ·6 min read
Ollama + Skales: The Complete Offline AI Stack
Install Ollama, pick the right model, connect to Skales, and verify you are fully offline. Plus: performance benchmarks and the best models for each task type.
- ·4 min read
Group Chat AI: What Happens When Multiple Models Debate Your Question
What if instead of asking one AI, you set multiple models with different perspectives arguing about your problem? Skales' Group Chat feature does exactly this.
- ·5 min read
How I Set Up an AI Assistant for My Parents (And They Actually Use It)
My parents are in their 70s. Setting up AI tools for them had failed three times before Skales. Here is what finally worked — and why.
- ·4 min read
Why Skales Doesn't Need Docker — And Why That Matters
Most open-source AI agent frameworks require Docker, Python, and a terminal. Skales requires none of these. Here's why that distinction matters.
- ·5 min read
Multi-Model AI Routing: Using the Right Model for Each Task
One model does not fit all tasks. Here is how to route queries intelligently — cheap models for simple tasks, expensive ones for complex — and cut your AI spend without sacrificing quality.
- ·6 min read
Creating Custom Skills for Your AI Agent: A Complete Guide
Custom skills are reusable prompt workflows with named parameters. Here is a complete guide to creating them: structure, parameters, examples, and advanced patterns.
- ·4 min read
Running Local AI on a Budget Laptop: A Practical Guide
You don't need a gaming PC or an Apple Silicon Mac to run local AI. Here's what actually works on modest hardware.
- ·5 min read
The Freelancer's AI Workflow: Proposals, Invoices, and Client Management
Freelancing means doing the work and running the business. Here is a complete AI-assisted workflow from first contact to paid invoice — with real time estimates.
- ·5 min read
The Privacy Problem with Every Major AI Tool — And How to Fix It
ChatGPT, Copilot, Gemini — every major AI tool has the same privacy problem. Here's what that problem is and the only way to genuinely solve it.
- ·5 min read
Why Therapists Need Local AI (And Why Cloud AI Is a Risk)
Therapists generate hours of documentation daily. AI can help — but not if it sends patient content to cloud servers. Here is the only compliant path.
- ·5 min read
How to Actually Manage Your Email with AI (Not Just Summarize It)
Summarising email is the shallow end of AI. The deep end is categorisation, auto-drafted replies, follow-up scheduling, and inbox patterns that actually change how email feels.
- ·5 min read
Why Healthcare Professionals Are Turning to Local AI
Patient data is among the most sensitive information that exists. Cloud AI tools and healthcare regulations are fundamentally incompatible. Local AI is the only compliant path.
- ·4 min read
From Prompt to App: The Rise of No-Code AI Development
In 2025, 'I can't code' stopped being a barrier. Here's what no-code AI development actually looks like in practice.
- ·6 min read
I Replaced ChatGPT with a Local AI Agent for a Month
Four weeks, one experiment: no ChatGPT, no Claude, no cloud AI. Only Skales with Ollama. Here is exactly what worked, what did not, and what the verdict is.
- ·4 min read
5 Things AI Agents Can Do That Chatbots Simply Can't
A chatbot answers questions. An agent takes actions. Here is the concrete difference — and why it changes what you can delegate.
- ·5 min read
Browser Automation with AI: Beyond Selenium and Puppeteer
Traditional browser automation breaks when the page changes. AI-driven automation understands pages semantically — here is what that difference means in practice.
- ·5 min read
My 6-Year-Old Built a Snake Game with AI in 12 Minutes
She wanted a game. I told her to ask Skales to build it. Twelve minutes later she was playing. Here is exactly what happened — and what it means.
- ·6 min read
How AI Tool Calling Works: From Prompt to Action
Function calling, tool use, the ReAct loop — here is exactly how AI agents go from a language model output to a real action in the world.
- ·4 min read
How Local AI Changes Remote Work — Permanently
Remote workers deal with asynchronous communication overload, timezone gaps, and the constant context-switching between tools. Local AI solves all three.
- ·6 min read
How AI Agents Help with ADHD and Executive Dysfunction
ADHD is not a deficit of intelligence. It is a difference in executive function — and AI agents target exactly the functions that are hardest: initiation, prioritisation, and working memory.
- ·6 min read
How AI Agent Memory Works: From Conversations to Long-Term Knowledge
Context windows, conversation history, bi-temporal facts, soul.json and human.json — here is how Skales builds persistent knowledge from your conversations.
- ·4 min read
Voice-First AI: Building Accessible Workflows for Everyone
For users with motor disabilities, dyslexia, or anyone who thinks faster than they type, voice-first AI is not a feature — it is the point.
- ·6 min read
The AI Agent Landscape in 2026: Who's Building What
From OpenAI's Operator to Anthropic's Claude Desktop to open-source frameworks — here is a fair overview of the AI agent landscape in 2026 and where it is heading.
- ·4 min read
AI for ADHD: How Automation Reduces Cognitive Overload
Task-switching, email overwhelm, and forgotten commitments are the daily battles of ADHD. Here is how AI automation changes that equation.
- ·7 min read
Building a Desktop AI Agent with Electron and Next.js
A technical deep-dive into the Skales architecture: why Electron, why Next.js as the UI framework, the IPC bridge, and the complete tool execution flow.
- ·5 min read
The Teacher's Guide to AI Tools That Don't Violate Student Privacy
EU schools cannot legally use most cloud AI tools with student data. Here's what compliant AI-assisted teaching actually looks like.
- ·6 min read
Getting Started with Ollama: Run AI Models on Your Own Machine
Ollama is the easiest way to run AI models locally. This complete beginner's guide covers installation, model selection, RAM requirements, and connecting it to Skales for a GUI.
- ·5 min read
Llama 3 vs GPT-4o: Which Model Should You Run Locally in 2026?
GPT-4o is the gold standard. But Llama 3.1 70B comes surprisingly close — and runs entirely on your hardware. Here's the honest comparison.
- ·5 min read
How to Build a Personal AI Workflow Without Giving Up Your Data
A practical guide to building a daily AI workflow where every sensitive task stays on your machine and only non-sensitive queries go to the cloud.
- ·6 min read
The Best AI Providers for Local Agents in 2026: Compared
OpenRouter, OpenAI, Anthropic, Google, Ollama — all work with local agents. Here is a no-nonsense comparison of cost, quality, privacy, and speed to help you choose.
- ·5 min read
AI Agent Security: What You Should Know Before Installing One
An AI agent with access to your files, email, and shell is a powerful thing. Here is an honest guide to the security questions you should ask before installing one.
- ·4 min read
Why I Stopped Paying for Copilot — And What I Use Instead
GitHub Copilot is a great product. But at $10-19/month with no offline mode and Microsoft telemetry, there are better options for privacy-conscious developers.
- ·5 min read
The Forgotten Users: Why Most AI Tools Fail on Accessibility
Screen reader support is broken. Keyboard navigation is incomplete. Voice-only users are largely ignored. AI tools have an accessibility problem — and the industry is not taking it seriously.
- ·5 min read
What Can AI Agents Actually Do in 2026? A Realistic Assessment
The demos are impressive. The marketing is breathless. Here is an honest assessment of what AI agents actually do well, what they still struggle with, and what is genuinely coming next.
- ·4 min read
Whisper Is the Best Thing to Happen to Voice AI Since Siri
OpenAI's Whisper transcription model runs locally, supports 99 languages, and beats every cloud STT service in accuracy. Here's why it matters.
- ·6 min read
Running AI Agents Without Docker: Yes, It's Possible
Most open-source AI agent frameworks require Docker, Python, and a terminal. This is a significant barrier. Here is what exists instead — and how the setup actually compares.
- ·5 min read
Desktop Automation in 2025: What AI Agents Can Actually Do
Beyond chatting — AI agents that read your email, control your browser, manage your calendar, and run on a schedule. Here is the current state of the art.
- ·6 min read
The Case for Local-First AI: Privacy, Cost, and Control
The local-first software movement has been building for a decade. Applying its principles to AI is not just about privacy — it is about what kind of relationship you want with the tools that shape your thinking.
- ·4 min read
The Real Cost of AI Subscriptions: An Honest Monthly Breakdown
ChatGPT Plus is $20/month. Copilot Pro is $20/month. Perplexity is $20/month. It adds up fast — here's what you're actually paying for.
- ·5 min read
Local AI vs Cloud AI: What's the Difference and Why It Matters
Cloud AI is convenient. Local AI is private. Here is exactly what the difference is, what each costs, and how to decide which one belongs in your workflow.
- ·4 min read
Ollama vs ChatGPT: The Case for Running AI on Your Own Hardware
ChatGPT is powerful. Ollama is private and free. Here's how they actually compare — and when local wins.
- ·5 min read
AI Agents Are Not Chatbots: Why the Distinction Matters
Most products marketed as 'AI agents' are chatbots with an extra tab. Here is what real agency requires — and a simple test for telling the difference.
- ·5 min read
GDPR and AI: Why Local Models Are the Only Safe Choice for European Users
GDPR compliance is not just about cookie banners. When AI tools send your data to US servers, you may already be violating European law.
- ·5 min read
What Is an AI Agent? A Simple Explanation for Non-Technical People
AI agents, chatbots, assistants — the terms get thrown around interchangeably. Here is a clear, jargon-free explanation of what an AI agent actually is.
- ·4 min read
Why Your AI Assistant Shouldn't Live in the Cloud
Every message you send to a cloud AI lands on someone else's server. Here's why that matters — and what you can do about it.