DDevn NieEditor3d ago · in AIEvaluating a RAG system without a golden dataset: the three numbers that still tell you something
MMarsEditor3d ago · in AICaching, batching and quantisation: the three server settings that decide throughput
WWister XueEditor4d ago · in AIAgent frameworks by what they refuse to do: CrewAI, AutoGen, smolagents, Agno
EEdenEditorOct 17 · in AIAn MCP server entry by hand: the config file, the transport, and the three ways it fails
NNicoEditorOct 17 · in AIA local model as a code reviewer: what it catches, and what it quietly misses
WWister XueEditorOct 16 · in AIPrompt caching, JSON mode and tool calls: three levers that cut an agent's bill without changing the model
KKaceyEditorOct 14 · in AISmall models for classification: a one-billion-parameter model doing eighty percent of the work
EEvieEditorOct 11 · in AIEmbedding models: when a bigger one is worth it, and how to tell before you index anything
WWister XueEditorOct 9 · in AILLM gateways: when does a small team need LiteLLM, and when is the SDK enough?
VValerieEditorOct 9 · in AIWhat "agent skills" repositories are, and why three of them sat on the weekly list for seven weeks
KKikiEditorOct 9 · in AIWhat re-checking vendor pricing pages found: Copilot dropped annual billing, Cursor's yearly price is not stated
TTerryEditorOct 8 · in AIA local model as a daily driver: Ollama + Open WebUI, Jan, GPT4All, LocalAI, and where LM Studio fits
HHysen YanEditorOct 8 · in AIThree AI coding tools changed hands this year: Windsurf to Devin, Tabnine to Tricentis, Continue to Cursor