The Radar
Monday, 6 April 2026
Today's picks
AutoKernel
AI InfrastructureOpen-source framework that applies autonomous LLM agents to GPU kernel optimisation for PyTorch models.
GPU kernel optimisation is one of the most specialised and painful parts of ML engineering. Having an LLM agent automatically write, test, and iterate on CUDA kernels could democratise performance optimisation that currently requires years of expertise.
MaxToki
AI ResearchAI foundation model that predicts how cells age and what interventions might slow the process.
Most biology models see cells as frozen snapshots, but MaxToki understands cellular dynamics over time. This temporal understanding could unlock genuine breakthroughs in ageing research rather than just pattern matching on static data.
Also on the radar
Cabinet
AI WritingThe knowledge management space is crowded, but the Paperclip reference suggests this might actually understand context across your entire knowledge base rather than just doing RAG over documents.
Apex Protocol
AI AgentsBuilding on Model Context Protocol for financial trading could standardise how AI agents interact with markets. Whether this democratises algorithmic trading or just creates more sophisticated ways to lose money remains to be seen.
Onepilot
AI CodingThe idea of managing production deployments from your phone sounds either brilliant or terrifying. If they've solved the security and interface challenges, this could genuinely change how we think about mobile development workflows.
TermHub
Developer ToolsAI agents need secure, controlled access to terminal operations without breaking everything. An open-source gateway that handles permissions and sandboxing could become essential infrastructure as agents get more autonomous.
Hacker News
Show HN: Cabinet – Kb+LLM (Like Paperclip+Obsidian)
16 pts 11 commentsA knowledge base system that combines LLM capabilities with note-taking, positioned as a hybrid of Microsoft's old Paperclip assistant and modern knowledge management tools like Obsidian.
Apex Protocol – An open MCP-based standard for AI agent trading
12 pts 4 commentsAn open standard built on Model Context Protocol that aims to standardise how AI agents interact with trading systems and financial markets.
Onepilot – Deploy AI coding agents to remote servers from your iPhone
9 pts 4 commentsA mobile application that allows developers to deploy and manage AI coding agents on remote servers directly from iOS devices.
Show HN: ACE – A dynamic benchmark measuring the cost to break AI agents
8 pts 3 commentsA security testing framework that measures how much it costs to successfully exploit or break AI agent systems, providing metrics for adversarial robustness.
AI agents promise to 'run the business,' but who is liable if things go wrong?
6 pts 0 commentsAnalysis of the legal and liability questions that arise as AI agents become more autonomous in business operations and decision-making processes.