The Radar
Sunday, 3 May 2026
Today's picks
KAME
AI InfrastructureReal-time speech-to-speech with injected LLM knowledge.
Sakana AI cracked the latency problem that kills most voice AI products. Their tandem architecture injects LLM smarts into speech without the usual delays, which could finally make AI conversations feel natural instead of awkward ping-pong matches.
Flue
AI AgentsTypeScript framework for building AI agents.
Finally, a proper TypeScript framework for agent development instead of cobbling together Python scripts and hoping they work. The sandbox approach could solve the reliability nightmares that plague most agent implementations.
Also on the radar
xAI Custom Voices
AI InfrastructurexAI's Custom Voices brings production-ready voice cloning to developers with minimal training data. The one-minute requirement makes this practical for real applications, not just demos.
Mljar Studio
Data ToolsLocal AI data analysis that outputs proper notebooks instead of black box results. Smart move keeping everything on-device for sensitive datasets whilst maintaining reproducible workflows.
MiMo-V2.5-Pro
AI CodingXiaomi's MiMo-V2.5-Pro delivers Claude Opus-level coding whilst being fully open-weight and burning 4x less compute. The autonomous multi-hour coding sessions could reshape how we think about AI pair programming.
Hacker News
Flue is a TypeScript framework for building the next generation of agents
89 pts 51 commentsA new TypeScript framework designed specifically for AI agent development. Promises better reliability and developer experience compared to existing Python-heavy tooling.
Show HN: Mljar Studio – local AI data analyst that saves analysis as notebooks
65 pts 10 commentsLocal AI-powered data analysis tool that maintains reproducible workflows by saving results as proper notebooks. Keeps sensitive data on-device whilst providing intelligent analysis.
Show HN: Large Scale Article Extract of Newspapers 1730s-1960s
46 pts 18 commentsMassive digitised newspaper archive spanning over two centuries. Useful dataset for historical AI research and natural language processing applications.
Preliminary Findings on AI Automation from Worker Evaluations
6 pts 0 commentsResearch paper examining how workers evaluate AI automation in their roles. Early findings on acceptance patterns and productivity impacts.
Show HN: Which public repos are friendliest to an AI coding agent?
5 pts 0 commentsTool that analyses GitHub repositories to determine how well they work with AI coding agents. Useful for developers building agent-assisted workflows.