JetBrains
5 storiesJetBrains is a software company known for IDEs and developer tools. Recent coverage has focused on AI coding: it released the Mellum2.1 small-scale MoE model with 12B total and 2.5B active parameters, described the Air Context RAG pipeline for semantic code search, and consolidated JetBrains Central, its CLI, cloud agents and governance into JetBrains Air. Its SkillsBench also found little savings from RTK's token-reduction claims.
Related topics
JetBrains released its Mellum2.1 model collection on the LocalLLaMA community, including a Thinking-GGUF version with 12B total parameters and 2.5B active parameters in a MoE architecture.
JetBrains engineers detail building Air Context, a RAG pipeline for semantic code search, covering parsing, chunking and vectorization. Traditional grep and keyword search require agents to know exact text in advance and fail on abstract queries; the pipeline instead gives LLM agents precise, citable evidence from repositories, improving efficiency on large codebases.
JetBrains is consolidating JetBrains Central, its CLI, shared context, cloud agents, automations and governance into JetBrains Air, covering the agentic development lifecycle beyond the IDE. It expects agentic development to be multi-vendor and says the IDE remains central to its plans.
Discussion on HN examines RTK (Rust Token Killer), which compresses terminal output to reduce tokens read by AI agents; despite README claims up to 90% reduction, JetBrains’ SkillsBench and independent benchmarks found little or no cost savings.
A Hacker News post urges individuals and organizations to resist current 'AI', arguing it is seductive and harmful until its significant harms are eliminated or compensated. The author avoids programming AIs, acknowledges some useful features, and cites data center power roughly equal to 100,000 households and water to 50,000 households.
That is everything