Copilot Instructions, Agents, and Skills: The Missing Control Layer
How custom instructions, AGENTS.md, and skills give your local Copilot the context it needs to stop making stupid mistakes.
EN
-
-
ENMCP Is How Local Copilot Becomes Useful
How MCP servers turn a local LLM into a real agent: test runners, linters, semantic search, and why tool access compensates for weaker reasoning.
-
ENHow I Set Up GitHub Copilot CLI on Local Hardware
The complete setup guide: vLLM and llama.cpp configs, environment variables, the dual-mode trick, debugging /v1/models, every failure mode, and why owning the runtime changes the equation.
-
ENLocal LLMs Under the Hood
Text becomes tokens, tokens become vectors, vectors consume memory, and memory pressure adds latency. A complete breakdown of the LLM inference pipeline.
-
ENMayday! Mayday! We're Running Out of Fuel!
Running out of tokens? Will you be part of this new form of slavery or will join the local ai rebellion? Your AI. Your rules.
-
ENMoving Toward Spec-Driven Development with OpenSpec or Spec Kit
AI can generate code faster than we can govern it. This article connects drift, specs, OpenSpec, and GitHub Spec Kit through the idea that really matters now: orchestration.
-
ENSpec-Driven Development: Controlling AI-Generated Drift
How specification-driven development serves as a control system to maintain product consistency when AI generates code at unprecedented speeds.
-
ENGitHub Copilot: What Consumes Premium Request
Practical guide to GitHub Copilot Premium Request Units (PRUs): what features consume them, what doesn't, and how to optimize usage when using Copilot autocomplete, chat, CLI, agents, and code reviews.
-
ENBuilding a Sovereign Home Network: Multi-Home, Cloudless, and Resilient
How to design and operate a fully self-owned, multi-home network with WireGuard, Docker, and Home Assistant—no cloud, no vendor lock-in. Covers edge VPS, overlay networking, split-horizon DNS, automation, security, and failure recovery.
-
ENRunning Claude Code Against Your Local GPU with Docker Model Runner
Bridging the API gap between Claude Code (Anthropic) and local models using a lightweight proxy. Complete guide with architecture diagrams and autostart configuration.