LSearch: a Windows Spotlight clone with embeddings and a fine-tuned 2x4B MoE
Scutoidzz · reddit · 2026-10-12
Developer scutoidzz released LSearch, a Spotlight-style search app for Windows with an embedding model and a local MoE built in.
- The MoE combines two Qwen 3.5 4B models — one fine-tuned for coding, one for agentic work; deliberately not shoehorned into everything, it's only used to fix code
- Embeddings run on embeddinggemma 2 (270M), no image support yet
- The app itself was written with Opus 5.5
- Model is on Hugging Face (scutoid/2x4B); with 16GB+ RAM the app offers to download it automatically
- Source and releases: https://github.com/blueprintnerd/LSearch
More from coding & agent
- Qwen releases QwenGyre: elastic GPU scheduling speeds long-horizon agent RL training by up to 1.85x — Alibaba_Qwen · 2026-10-12
- Logit-based decision models batch naturally: routing, scoring, yes/no in one forward pass — LysandreJik · 2026-10-12
- Hugging Face's transformers serve can turn 250 LLMs & VLMs into logit-based decision models — LysandreJik · 2026-10-12
- How One Dev Rents a 64-Core, 256GB Box for $800/Month to Run 30 Parallel AI Subagents — TheZachMueller · 2026-10-12
- OAuth isn't enough for agents: Keycard demos short-lived creds and task-based policy — AI Engineer · 2026-10-12
- NYC coffee shop spot check: only 6 of 22 laptops used AI, all ChatGPT — thedealdirector · 2026-10-12