AI

Reasonix stores global state in ~/.reasonix/, project state inside the project, and the runtime in two more places. Knowing which directory holds what makes backup scripts trivial. The trickiest file is .env — it is excluded from MCP child environments but you still want it under version control, just redacted. A backup script #!/usr/bin/env bash... » read more

Reasonix already speaks the Model Context Protocol. Adding an MCP server means one more [[plugins]] entry under ~/.reasonix/config.toml (or reasonix.toml per-project). Three transports are supported, and MCP prompts automatically become slash commands. The configuration Two files: TOML for non-secret configuration, .env for key values. Resolution order is flag > ./reasonix.toml > ~/.reasonix/config.toml > built-in defaults.... » read more

A Reasonix session that runs for two hours will eventually forget what file you were in, write something into the wrong directory, or charge you for repeating the same prefix. The slash commands are the controls that keep the cache warm and the diff readable. Most of them are local — they cost nothing and... » read more

Reasonix is hard-coded to talk to a provider endpoint, not to a model name. Adding a new endpoint is a [[providers]] entry in ~/.reasonix/config.toml and a matching API key in ~/.reasonix/.env. The same shape works for local llama.cpp, Ollama, vLLM, and remote gateways. Common endpoints Provider base_url Sample model API key env Ollama (local) http://127.0.0.1:11434/v1... » read more

The VS Code extension is a thin shell. It does not bundle the Reasonix CLI; it starts your local reasonix acp backend and adds chat, workspace sessions, editor context, and approval cards. If reasonix --version works on the terminal, the extension will work; if it does not, the extension cannot save you. Before you start... » read more

Reasonix is a DeepSeek-native coding agent that runs as a single Go binary in your terminal. It keeps DeepSeek’s prompt prefix stable so long sessions stay cheap. On Ubuntu you can install it three ways; pick the one that matches what your machine already has. Before you start You need three things on a stock... » read more

Why three frameworks, and why now If you want to run a neural network on a phone, a Raspberry Pi, a smart toaster, or a laptop without a beefy GPU, you used to have one reasonable choice: TensorFlow Lite. In 2026 you have three serious ones — ExecuTorch, Apache TVM, and LiteRT (the renamed TensorFlow... » read more

To be honest, I’ve always had one regret: Linux has so many excellent terminal AI agents, but what about GUI applications? My parents don’t know how to open a terminal. My colleagues don’t want to touch the command line to write documents. Don’t Linux desktop users deserve a nice-looking AI chat window? —Of course they... » read more

Have you ever had this experience? You open a terminal, facing a wall of error messages, not knowing where to start. Copy and paste into ChatGPT, switch back to the terminal to execute, nope — switch back to ask again, switch back… I’ve had enough of that. So this weekend, I spent two days testing... » read more

15 Lines of Code to Teach AI to Turn Lights On and Off: A Smart Energy-Saving Butler with Neural Networks No rules, no if-else, just “trial and error + reward” — let AI learn to intelligently control lighting. This is the magic of Reinforcement Learning! Problem: When Should the Light Be On? Is your home... » read more

Teach an AI Beginner to Define Their Own Model and Implement Image Classification in 30 Minutes Have you ever wondered: “What does an AI model actually look like? Can I write one myself?” “Online tutorials are too complex, all code, totally incomprehensible” “I want to learn AI but don’t know where to start” Don’t worry!... » read more

The Essence of AI Models: Using Parameters to Fit High-Dimensional Functions Humans Can Hardly Understand In today’s AI boom, we often hear terms like “large models,” “parameter count,” and “training.” But do you truly understand: the essence of AI models is using a bunch of parameters to fit a set of high-dimensional functions that humans... » read more