GoKawiil
Tech News
A new tool called Pensieve was benchmarked against plain grep-based search inside Claude Code across four real repositories ranging from 500 to 35,000 files. In held-out code-finding tasks, Pensieve located the correct file in 131 of 144 runs versus 123 of 144 for grep, while using 43% fewer tool calls, costing 18% less per lookup, and finishing 21% faster, with each search taking 0.1 to 0.2 seconds.
Cactus launched Needle 3, a foundation model small enough to run as a single 8-29MB binary on phones, wearables, robots, smart home hubs and cars. Built on what the company calls a Simple Attention Network, it sacrifices general chat ability to specialize in three tasks: text embedding for local search, structured data extraction from messy text, and tool calling that maps spoken requests to app functions. The company claims it outperforms models ten times larger on mobile tool calls and matches models two to three times bigger on extraction.