Rost Glukhov
LightNews LightNews
rosgluk.bsky.social
Rost Glukhov
@rosgluk.bsky.social
60 followers 22 following 580 posts
Software developer, musician and photographer
living in Melbourne, Australia

https://www.glukhov.org/
https://glukhov.au/
Posts Replies Media Videos
rosgluk.bsky.social
gstack: AI Software Engineering Stack

#AI Coding #AI Agents #Open Source #LLM #AI #Dev

https://www.glukhov.org/ai-devtools/gstack/
gstack: AI Software Engineering Stack
Explore gstack, an opinionated AI engineering stack for Claude Code: its skills, install steps, and how it combines with Superpowers, OpenSpec, and Spec Kit.
www.glukhov.org
September 26, 2026 at 2:33 AM
rosgluk.bsky.social
LLM Systems: Operational Reference for Self-Hosted Inference

#AICoding #LLM #AI #Dev #OpenSource #SelfHosted #llama.cpp #GGUF

https://llmsystems.dev/
LLM Systems: Operational Reference for Self-Hosted Inference
llmsystems.dev is live: curated model and GGUF artifact records, llama.cpp, hardware memory classes, and weights-only VRAM bands with explicit evidence labels (SOURCE, POLICY, OBSERVED). Built for people who self-host — not a blog, not fake fit claims.
llmsystems.dev
September 25, 2026 at 8:30 AM
rosgluk.bsky.social
Self-Hosted Deep Research Systems: 12 Tools Compared

#AI Agents #RAG #LLM #AI #Self-Hosting #SelfHosting #Open Source #Ollama

https://www.glukhov.org/ai-systems/comparisons/deep-research-with-ai/
Self-Hosted Deep Research Systems: 12 Tools Compared
Compare 12 self-hosted Deep Research systems: GPT Researcher, Onyx, Open WebUI, Khoj, Vane and more. Architectures, local LLM, RAG, licenses.
www.glukhov.org
September 24, 2026 at 10:12 AM
rosgluk.bsky.social
The Efficient Frontier of Open Models: Finding the Sweet Spot in 2026

#LLM #Open Source #NVidia

https://www.glukhov.org/llm-performance/benchmarks/efficient-frontier-of-open-models-2026/
The Efficient Frontier of Open Models: Finding the Sweet Spot in 2026
In 2026 the open-model efficient frontier sits at 25-34B: Qwen 3.8 27B and Llama 4 Scout deliver near-frontier capability at a fraction of 70B+ cost.
www.glukhov.org
September 22, 2026 at 10:01 AM
rosgluk.bsky.social
Mnemosyne for Hermes Agent: Local Memory Quickstart

#Hermes #AI #LLM #Architecture #SelfHosting #Python #SQL

https://www.glukhov.org/ai-systems/hermes/mnemosyne-memory/
Mnemosyne for Hermes Agent: Local Memory Quickstart
Install Mnemosyne as a local Hermes memory provider, configure recall and retention, and inspect the SQLite store.
www.glukhov.org
September 19, 2026 at 9:49 AM
rosgluk.bsky.social
Self-Reinforcing Memory Loops in AI Agents: Causes and Fixes

#Hermes #OpenClaw #AI #LLM #Architecture #SelfHosting #Security

https://www.glukhov.org/ai-systems/memory/self-reinforcing-memory-loops/
Self-Reinforcing Memory Loops in AI Agents: Causes and Fixes
Agent memory can turn model inferences into future evidence. How those loops form, why they matter, and which write-path controls limit them.
www.glukhov.org
September 18, 2026 at 9:11 AM
rosgluk.bsky.social
OpenSpec Rejected Proposals: A Decision Memory Convention

#AI Coding #LLM #AI #Dev #Open Source #Architecture

https://www.glukhov.org/ai-devtools/openspec/handling-rejected-proposals/
OpenSpec Rejected Proposals: A Decision Memory Convention
OpenSpec has no native rejected-proposal state. Learn why, what the maintainers decided, and the decision.md convention that stops agents reopening settled investigations.
www.glukhov.org
September 17, 2026 at 11:13 AM
rosgluk.bsky.social
OpenSpec Quickstart: Install, Workflow, and Common Pitfalls

#AI Coding #LLM #AI #Dev #Open Source #Architecture #Node.js

https://www.glukhov.org/ai-devtools/openspec/
OpenSpec Quickstart: Install, Workflow, and Common Pitfalls
Install OpenSpec, the change-centric spec-driven development CLI. Learn the explore-propose-apply-archive loop, brownfield setup, and where it breaks.
www.glukhov.org
September 16, 2026 at 8:44 AM
rosgluk.bsky.social
How to Migrate from OpenClaw to Hermes Agent Safely

#Hermes #OpenClaw #AI #LLM #AI-Agents #DevOps #SelfHosting

https://www.glukhov.org/ai-systems/hermes/migrate-from-openclaw/
How to Migrate from OpenClaw to Hermes Agent Safely
Migrate OpenClaw to Hermes Agent with dry-run checks, precise memory and skill mapping, secret handling, channel cutover, rollback, and manual fixes.
www.glukhov.org
September 15, 2026 at 12:00 PM
rosgluk.bsky.social
LLM Hosting in 2026: Local, Self-Hosted and Cloud Infrastructure Compared

#AI #LLM #hosting #Self-Hosting #ollama #llama.cpp #vllm #infrastructure

https://www.glukhov.org/llm-hosting/
LLM Hosting in 2026: Local, Self-Hosted and Cloud Infrastructure Compared
Complete guide to LLM hosting in 2026. Compare Ollama, llama.cpp, vLLM, TGI, Docker Model Runner, LocalAI and cloud providers. Learn cost, performance, and infrastructure trade-offs.
www.glukhov.org
September 14, 2026 at 10:13 AM
rosgluk.bsky.social
llama.cpp vs Ollama in 2026: Which Runtime Should You Run?

#llamacpp #Ollama #LLM #SelfHosting #Self-Hosting #API

https://www.glukhov.org/llm-hosting/comparisons/llama-cpp-vs-ollama/
llama.cpp vs Ollama in 2026: Which Runtime Should You Run?
Compare llama-server and Ollama for local LLM hosting in 2026: GGUF management, APIs, GPU control, KV cache, model lifetime, and migration triggers.
www.glukhov.org
September 12, 2026 at 10:15 AM
rosgluk.bsky.social
ROCm vs Vulkan for AMD Local LLM Hosting: 2026 Guide

#LLM #SelfHosting #llamacpp #Ollama #vLLM #Docker #Linux

https://www.glukhov.org/llm-hosting/comparisons/amd-rocm-vs-vulkan-llm-hosting/
ROCm vs Vulkan for AMD Local LLM Hosting: 2026 Guide
Compare AMD ROCm and Vulkan backends for llama.cpp, Ollama, LM Studio, vLLM and TGI, with build commands, verification checks and a practical 2026 verdict.
www.glukhov.org
September 11, 2026 at 9:24 AM
rosgluk.bsky.social
KV Cache on 16 GB GPUs: Making Long Context Actually Fit

#LLM #llamacpp #vLLM #Ollama #GPU #Self-Hosting #SelfHosting #NVidia #Hardware

https://www.glukhov.org/llm-performance/optimization/kv-cache-16gb-long-context/
KV Cache on 16 GB GPUs: Making Long Context Actually Fit
Fit 32K to 128K LLM context into 16 GB VRAM by calculating KV cache cost, choosing cache precision, and tuning llama.cpp, vLLM, or Ollama safely.
www.glukhov.org
September 9, 2026 at 10:55 AM
rosgluk.bsky.social
What Comes After LLMs? Mamba, Diffusion & World Models

#AI #LLM #Architecture #DeepLearning

https://www.glukhov.org/ai-models/architectures/what-comes-after-llms/
What Comes After LLMs? Mamba, Diffusion & World Models
Transformers are hitting walls in compute, data, and energy. State space models (Mamba), diffusion language models, and JEPA world models may come next.
www.glukhov.org
September 8, 2026 at 10:42 AM
rosgluk.bsky.social
Backup and Restore Gitea server

#linux #self-hosting #git #gitea #backup #DevOps

https://www.glukhov.org/developer-tools/git-and-forges/gitea-backup-restore/
Backup and Restore Gitea server
Backup and restore a self-hosted Gitea server with the built-in gitea dump command, plus Docker steps, restore-repo, and troubleshooting tips.
www.glukhov.org
September 7, 2026 at 9:42 AM
rosgluk.bsky.social
Agent Skills vs MCP Servers: Decision Framework

#AI #Hermes #OpenClaw #AI Coding

https://www.glukhov.org/ai-systems/mcp/agent-skills-vs-mcp-servers-decision-framework/
Agent Skills vs MCP Servers: Decision Framework
Choose between Agent Skills and MCP servers using connectivity, credentials, context cost, security, and a practical thin-server pattern for AI agents.
www.glukhov.org
September 6, 2026 at 1:30 AM
rosgluk.bsky.social
Ollama to vLLM: When to Migrate Your Local LLM Server

#Ollama #vLLM #LLM #AI #Self-Hosting #Docker #API #DevOps

https://www.glukhov.org/llm-hosting/comparisons/ollama-to-vllm-migration/
Ollama to vLLM: When to Migrate Your Local LLM Server
Learn when to migrate from Ollama to vLLM. Migration signals, planning steps, Docker Compose setup, and a practical checklist for moving your local LLM server.
www.glukhov.org
September 5, 2026 at 6:45 AM
rosgluk.bsky.social
Superpowers Quickstart: Install, Workflow, and Tryout

#AI Coding #LLM #AI #Dev #Open Source #Architecture

https://www.glukhov.org/ai-devtools/superpowers/
Superpowers Quickstart: Install, Workflow, and Tryout
Install Superpowers, the open-source skills framework that enforces brainstorm-to-TDD spec-driven development across Claude Code, Cursor, and other agents.
www.glukhov.org
September 3, 2026 at 11:34 AM
rosgluk.bsky.social
OpenCode CLI in Practice: Workflows, Automation, and Pitfalls

#OpenCode #Cheatsheet #AI Coding #AI #LLM #Dev #DevOps

https://www.glukhov.org/ai-devtools/opencode/opencode-cli-in-practice/
OpenCode CLI in Practice: Workflows, Automation, and Pitfalls
A practical guide to OpenCode from the command line: opencode run, scripting, CI automation, agents, permissions, local models, and real failure modes.
www.glukhov.org
August 29, 2026 at 7:43 AM
rosgluk.bsky.social
GPUs for AI in 2026: NVIDIA, AMD, Intel Compared

#GPU #AI #NVIDIA #Hardware #infrastructure #LLM #SelfHosting

https://www.glukhov.org/hardware/ai/gpu-comparison-ai-workloads-2026-nvidia-amd-intel/
GPUs for AI in 2026: NVIDIA, AMD, Intel Compared
Compare NVIDIA Blackwell, AMD Radeon AI Pro R9700, and Intel Arc Pro B70 for local LLM inference. VRAM, bandwidth, software ecosystem, and real-world recommendations.
www.glukhov.org
August 26, 2026 at 8:40 AM
rosgluk.bsky.social
LLM Hosting in 2026: Local, Self-Hosted and Cloud Infrastructure Compared

#AI #LLM #hosting #Self-Hosting #ollama #llama.cpp #vllm #infrastructure

https://www.glukhov.org/llm-hosting/
LLM Hosting in 2026: Local, Self-Hosted and Cloud Infrastructure Compared
Complete guide to LLM hosting in 2026. Compare Ollama, llama.cpp, vLLM, TGI, Docker Model Runner, LocalAI and cloud providers. Learn cost, performance, and infrastructure trade-offs.
www.glukhov.org
August 25, 2026 at 10:49 AM
rosgluk.bsky.social
AI Developer Tools: The Complete Guide to AI-Powered Development

#AI #AI Coding #dev #llm #coding #devops #OpenCode

https://www.glukhov.org/ai-devtools/
AI Developer Tools: The Complete Guide to AI-Powered Development
Explore the modern AI developer tools ecosystem: AI coding assistants, GitHub Copilot, Claude Code, OpenCode, DevOps automation, GitOps, VS Code workflows, GitHub Actions, and programming language trends.
www.glukhov.org
August 24, 2026 at 11:30 AM
rosgluk.bsky.social
Superpowers Quickstart: Install, Workflow, and Tryout

#AI Coding #LLM #AI #Dev #Open Source #Architecture

https://www.glukhov.org/ai-devtools/superpowers/
Superpowers Quickstart: Install, Workflow, and Tryout
Install Superpowers, the open-source skills framework that enforces brainstorm-to-TDD spec-driven development across Claude Code, Cursor, and other agents.
www.glukhov.org
August 23, 2026 at 10:52 AM
rosgluk.bsky.social
Pi Coding Agent Review: Minimal, Hackable AI Coding CLI

#AI Coding #LLM #AI #Dev #Open Source #Security

https://www.glukhov.org/ai-devtools/pi/pi-coding-agent-review/
Pi Coding Agent Review: Minimal, Hackable AI Coding CLI
Pi Coding Agent review with quickstart, security notes, model support, extensions, sessions, and a practical verdict on this open-source terminal agent.
www.glukhov.org
August 22, 2026 at 5:28 AM
rosgluk.bsky.social
GPUs for AI in 2026: NVIDIA, AMD, Intel Compared

#GPU #AI #NVIDIA #Hardware #infrastructure #LLM #SelfHosting

https://www.glukhov.org/hardware/ai/gpu-comparison-ai-workloads-2026-nvidia-amd-intel/
GPUs for AI in 2026: NVIDIA, AMD, Intel Compared
Compare NVIDIA Blackwell, AMD Radeon AI Pro R9700, and Intel Arc Pro B70 for local LLM inference. VRAM, bandwidth, software ecosystem, and real-world recommendations.
www.glukhov.org
August 8, 2026 at 10:03 AM