PinnedRost Glukhov·Oct 14Qwen 4 Is Coming — and It Could Be the Biggest Leap for Local AI in 2026Alibaba’s next Qwen generation is in training. Historical benchmarks, an architecture preview, and release rumors suggest interesting…A response icon6A response icon6
PinnedInPractical LLM SystemsbyRost Glukhov·Oct 712 Self-Hosted Deep Research Systems ComparedDeep Research has become its own category of software, not just a model pointed at a search box. This article compares twelve self-hosted…A response icon2A response icon2
PinnedInPractical LLM SystemsbyRost Glukhov·Aug 23I Compared Every Major GPU for Local AI in 2026 — The Winner Wasn’t the Most ExpensiveYou can spend $1,800 on a GPU and still be unable to run the model you want. Or you can spend $950 and run it comfortably. The difference…A response icon32A response icon32
InPractical LLM SystemsbyRost Glukhov·14h agoHow to Pick an Uncensored LLM: Abliterated vs Fine-tunedPick an uncensored model: cost, fit, tool calls.
InPractical LLM SystemsbyRost Glukhov·3d agoTiered Context: A Knowledge Hierarchy for AI AgentsFive-layer knowledge hierarchy for agent memory.A response icon1A response icon1
Rost Glukhov·3d agoTrending Open Source AI ProjectsOpen source AI, grouped by the jobA response icon1A response icon1
InPractical LLM SystemsbyRost Glukhov·Nov 4Hindsight Memory for OpenCode, Cursor, and Other Coding AgentsProject memory that survives between sessions.
Rost Glukhov·Oct 29The RTX 5090, Mac Studio, DGX Spark and other Prices and Options Checkalso Intel Arc Pro B70, Radeon AI Pro R9700 -> in US, UK and Australia.A response icon1A response icon1
InGo SystemsbyRost Glukhov·Oct 24Bounding Concurrency in Go: errgroup, Worker Pools, and BackpressureCap the work in flight, not just cancel it.
InPractical LLM SystemsbyRost Glukhov·Oct 18Setting Up Hindsight Memory for Hermes — The Right WayA long context window is not memory. If you run Hermes Agent for more than one session, you need a store that keeps decisions, facts, and…
